Blue Marble release

.gitattributes

@@ -33,3 +33,9 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+assets/radar.png filter=lfs diff=lfs merge=lfs -text
+pico/model_qcs8275.dlc filter=lfs diff=lfs merge=lfs -text
+nano/model_qcs8275.dlc filter=lfs diff=lfs merge=lfs -text
+tiny/model_qcs8275.dlc filter=lfs diff=lfs merge=lfs -text
+small/model_qcs8275.dlc filter=lfs diff=lfs merge=lfs -text
+medium/model_qcs8275.dlc filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,112 @@
+---
+license: apache-2.0
+library_name: onnx
+pipeline_tag: object-detection
+tags: [earth-observation, remote-sensing, sentinel-2, satellite, edge-ai, onnx, int8, ships]
+---
+
+<div align="center">
+
+# Blue Marble Ships
+
+<img alt="License" src="https://img.shields.io/badge/License-Apache_2.0-2a78d6"/> <img alt="Sizes" src="https://img.shields.io/badge/Sizes-pico_·_nano_·_tiny_·_small_·_medium-2a78d6"/> <img alt="Device" src="https://img.shields.io/badge/Profiled_on-Arduino_VENTUNO_Q_NPU-2a78d6"/> <img alt="Params" src="https://img.shields.io/badge/All_models-under_15M_params-2a78d6"/>
+
+</div>
+
+## Introduction
+
+Blue Marble Ships is a family of five ship detection models for Sentinel-2 imagery, from 32k to 13.5M parameters. Every model is under 15M parameters and runs fully in int8 on the Qualcomm Dragonwing QCS8275 NPU (Arduino VENTUNO Q).
+
+<p align="center"><img src="assets/benchmark.png" width="95%"></p>
+
+## Highlights
+
+- **86.6 AP** on our Sentinel-2 vessel test split (tiles unseen in training), **82.4 AP** on April 2026 scenes; ship length mean absolute error 17 m (medium).
+- **81.2 AP at 32k parameters**; ship position and length in one pass.
+- **3.6 ms** per 256x256 tile (medium) on the Arduino VENTUNO Q NPU, int8, every layer on the NPU.
+
+## Model Downloads
+
+| Model | Params | int8 size | VENTUNO Q latency | Ops on NPU | Ships test (ours): fp32 / int8 | Download |
+|:---|---:|---:|---:|:---:|:---:|:---|
+| Blue-Marble-Ships-pico | 0.032M | 0.14 MB | 1.0 ms | 62/62 | 81.2 / 78.7 | [`pico/`](./pico) |
+| Blue-Marble-Ships-nano | 0.19M | 0.36 MB | 1.0 ms | 76/76 | 84.3 / 81.6 | [`nano/`](./nano) |
+| Blue-Marble-Ships-tiny | 1.26M | 1.62 MB | 1.3 ms | 117/117 | 82.7 / 82.0 | [`tiny/`](./tiny) |
+| Blue-Marble-Ships-small | 2.4M | 2.86 MB | 1.5 ms | 130/130 | 80.8 / 78.3 | [`small/`](./small) |
+| Blue-Marble-Ships-medium | 13.5M | 14.58 MB | 3.6 ms | 224/224 | 86.6 / 86.2 | [`medium/`](./medium) |
+
+Each size folder contains:
+
+| File | Use |
+|:---|:---|
+| `model_qcs8275.dlc` | Compiled for the Qualcomm QCS8275 NPU (QNN runtime): runs as-is on the Arduino VENTUNO Q |
+| `model_int8.onnx` | int8 ONNX, the source of the compiled model; ONNX Runtime or your own NPU toolchain |
+| `model_fp32.onnx` | Full precision ONNX (CPU / GPU); the scores below are for this model |
+| `model.safetensors` + `config.json` | PyTorch weights for fine-tuning, with `models.py` in this repository |
+| `ventuno_q_profile.json` | On-device profile on the Arduino VENTUNO Q (Qualcomm AI Hub): the compute unit of every op, total latency, memory, load times |
+
+Scores in this table are for the exact files in this repository, re-scored on our held-out ships test split (332 ships). Latency and ops on NPU: on-device profile of `model_qcs8275.dlc` on an Arduino VENTUNO Q (Qualcomm AI Hub).
+
+## Evaluation Results
+
+### Models under 15M parameters
+
+| Model | Params | Ships test (ours) (AP) | Ships Apr-2026 (ours) (AP) |
+|:---|---:|---:|---:|
+| **Blue Marble pico** | 0.032M | 81.2 | 81.7 ± 0.6 |
+| **Blue Marble nano** | 0.19M | 84.3 | 80.1 ± 0.5 |
+| **Blue Marble tiny** | 1.26M | 82.7 | 80.7 ± 0.7 |
+| **Blue Marble small** | 2.4M | 80.8 | 82.2 ± 0.5 |
+| **Blue Marble medium** | 13.5M | **86.6** | **82.4 ± 0.8** |
+
+Blue Marble, test: the released checkpoint in this repository; Apr-2026: mean ± std over 3 training runs. Both are our own splits, not public leaderboards. Other models: published numbers, with inputs noted where they differ from ours (our models use Sentinel-2 only); * = trained and evaluated by us on the same data and split (parameter count includes our decoder). Best per column among models under 15M in bold. Labels: Global Fishing Watch Sentinel-2 vessel detections (AIS-matched, plus high-confidence detections without AIS); ambiguous detections are ignored. Test (332 ships) and Apr-2026 (April 2026 scenes) use tiles not seen in training. Ship length error: mean absolute error of the predicted length for matched ships. AP: detections are local maxima of the heatmap, a detection is correct if its centre is within 50 m of an unmatched label, all-point interpolated average precision.
+
+<p align="center"><img src="assets/radar.png" width="62%"></p>
+<p align="center"><img src="assets/size_vs_accuracy.png" width="70%"></p>
+<p align="center"><img src="assets/latency.png" width="70%"></p>
+
+## Quickstart
+
+```python
+import numpy as np, onnxruntime as ort
+
+MEAN = np.array([2357.089, 2137.385, 2018.788, 2082.986, 2295.651, 2854.537, 3122.849, 3040.560, 3306.481, 1473.847, 506.070, 2472.825, 1838.929], np.float32)[:, None, None]
+STD  = np.array([1624.683, 1675.806, 1557.708, 1833.702, 1823.738, 1733.977, 1732.131, 1679.732, 1727.26, 1024.687, 442.165, 1331.411, 1160.419], np.float32)[:, None, None]
+
+def prepare(dn):                       # dn: Sentinel-2 L1C digital numbers, [13, H, W], H and W multiples of 16
+    dn = np.clip(dn - 1000, 0, None)   # only for products from 2022 on (processing baseline 04.00+)
+    return ((dn - MEAN) / STD)[None].astype(np.float32)
+
+sess = ort.InferenceSession("tiny/model_int8.onnx")
+out = sess.run(None, {"image": prepare(dn)})[0]
+```
+
+PyTorch:
+
+```python
+import json, torch
+from safetensors.torch import load_file
+from models import BlueMarble                     # models.py from this repository
+
+cfg = json.load(open("tiny/config.json"))
+model = BlueMarble(cfg["arch"], cfg["num_classes"]).eval()
+model.load_state_dict(load_file("tiny/model.safetensors"))
+```
+
+- **Input:** 13 Sentinel-2 L1C bands, B01 to B12 including B8A and B10. Missing bands (other sensors): set to 0 after normalisation.
+- **Output:** `[1, 2, H, W]`: channel 0 ship-centre heatmap logit (sigmoid; 5x5 local maxima above 0.3 are ships), channel 1 log ship length (`length_m = exp(v) * 10`).
+
+## License
+
+Apache-2.0. Contains modified Copernicus Sentinel data.
+
+## Citation
+
+```bibtex
+@misc{bluemarble2026,
+  title  = {Blue Marble: Edge Earth Observation Models},
+  author = {Harsh Kumar},
+  year   = {2026},
+  url    = {https://huggingface.co/Civimatica}
+}
+```

medium/config.json

@@ -0,0 +1 @@
+{"arch": "medium", "num_classes": 2}

medium/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f6f2f07e1313410e18a003662392616fd8cfb3062a344ba4ebe351019878d40
+size 54163592

medium/model_fp32.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af935e7e1586d715a91223ae861fe01bb5c8cb9fdc6190e3d487273c3c44eb3a
+size 53817340

medium/model_int8.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b8cf6e2c3dff5f2e27ec9925ba98f2b68e92e717c97f3592bcc29ef870eb6e7a
+size 14578660

medium/model_qcs8275.dlc

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:04c68a09d66b7ba5594417dd70b125b984f8fb5c07fcb2fcd608b927ab261d4d
+size 16694992

medium/ventuno_q_profile.json

@@ -0,0 +1,1156 @@
+{
+ "device": "Arduino VENTUNO Q (Qualcomm Dragonwing QCS8275)",
+ "device_attributes": [
+  "os:ubuntu",
+  "chipset:qualcomm-qcs8275",
+  "chipset:qcs8275",
+  "chipset:qualcomm-dragonwing-iq-8275",
+  "hexagon:v75",
+  "soc-model:82",
+  "framework:qnn"
+ ],
+ "source": "Qualcomm AI Hub on-device profile job",
+ "profiled_at_utc": "2026-11-06 14:12:55",
+ "file": "model_qcs8275.dlc",
+ "precision": "int8",
+ "input": [
+  1,
+  13,
+  256,
+  256
+ ],
+ "ops_total": 224,
+ "ops_per_compute_unit": {
+  "NPU": 224,
+  "GPU": 0,
+  "CPU": 0
+ },
+ "inference_time_us": 3611,
+ "inference_peak_memory_bytes": 12673024,
+ "first_load_time_us": 4424796,
+ "warm_load_time_us": 176818,
+ "note": "peak memory is runtime working memory (activations and buffers), not the weight file size",
+ "on_device_vs_host_agreement": null,
+ "ops": [
+  {
+   "name": "Input",
+   "type": "Input",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "QNN_DATATYPE_FLOAT_32_converted_image_QNN_DATATYPE_UFIXED_POINT_8",
+   "type": "QNN",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "image_0231",
+   "type": "image",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stem/spectral/Conv",
+   "type": "/encoder/stem/spectral/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stem/spatial/spatial.0/Conv",
+   "type": "/encoder/stem/spatial/spatial.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stem/fuse/fuse.0/Conv",
+   "type": "/encoder/stem/fuse/fuse.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat0/lat0.0/Conv",
+   "type": "/decoder/lat0/lat0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.0/stages.0.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.0/stages.0.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.0/stages.0.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.0/stages.0.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.0/stages.0.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.0/stages.0.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.1/Add",
+   "type": "/encoder/stages.0/stages.0.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat.0/lat.0.0/Conv",
+   "type": "/decoder/lat.0/lat.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.1/stages.1.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.1/stages.1.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.1/stages.1.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.1/stages.1.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.1/stages.1.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.1/stages.1.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.1/Add",
+   "type": "/encoder/stages.1/stages.1.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.2/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.1/stages.1.2/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.2/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.1/stages.1.2/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.2/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.1/stages.1.2/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.2/Add",
+   "type": "/encoder/stages.1/stages.1.2/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat.1/lat.1.0/Conv",
+   "type": "/decoder/lat.1/lat.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.2/ReduceMean",
+   "type": "/encoder/stages.2/stages.2.1/net/net.2/ReduceMean",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.2/f/f.0/Conv",
+   "type": "/encoder/stages.2/stages.2.1/net/net.2/f/f.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.2/f/f.2/Conv",
+   "type": "/encoder/stages.2/stages.2.1/net/net.2/f/f.2/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.2/f/f.3/HardSigmoid",
+   "type": "/encoder/stages.2/stages.2.1/net/net.2/f/f.3/HardSigmoid",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.2/Mul",
+   "type": "/encoder/stages.2/stages.2.1/net/net.2/Mul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/Add",
+   "type": "/encoder/stages.2/stages.2.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.2/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.2/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.2/ReduceMean",
+   "type": "/encoder/stages.2/stages.2.2/net/net.2/ReduceMean",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.2/f/f.0/Conv",
+   "type": "/encoder/stages.2/stages.2.2/net/net.2/f/f.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.2/f/f.2/Conv",
+   "type": "/encoder/stages.2/stages.2.2/net/net.2/f/f.2/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.2/f/f.3/HardSigmoid",
+   "type": "/encoder/stages.2/stages.2.2/net/net.2/f/f.3/HardSigmoid",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.2/Mul",
+   "type": "/encoder/stages.2/stages.2.2/net/net.2/Mul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.2/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/Add",
+   "type": "/encoder/stages.2/stages.2.2/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.3/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.3/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.2/ReduceMean",
+   "type": "/encoder/stages.2/stages.2.3/net/net.2/ReduceMean",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.2/f/f.0/Conv",
+   "type": "/encoder/stages.2/stages.2.3/net/net.2/f/f.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.2/f/f.2/Conv",
+   "type": "/encoder/stages.2/stages.2.3/net/net.2/f/f.2/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.2/f/f.3/HardSigmoid",
+   "type": "/encoder/stages.2/stages.2.3/net/net.2/f/f.3/HardSigmoid",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.2/Mul",
+   "type": "/encoder/stages.2/stages.2.3/net/net.2/Mul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.3/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/Add",
+   "type": "/encoder/stages.2/stages.2.3/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.4/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.4/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.4/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.4/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.4/net/net.2/ReduceMean",
+   "type": "/encoder/stages.2/stages.2.4/net/net.2/ReduceMean",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.4/net/net.2/f/f.0/Conv",
+   "type": "/encoder/stages.2/stages.2.4/net/net.2/f/f.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.4/net/net.2/f/f.2/Conv",
+   "type": "/encoder/stages.2/stages.2.4/net/net.2/f/f.2/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.4/net/net.2/f/f.3/HardSigmoid",
+   "type": "/encoder/stages.2/stages.2.4/net/net.2/f/f.3/HardSigmoid",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.4/net/net.2/Mul",
+   "type": "/encoder/stages.2/stages.2.4/net/net.2/Mul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.4/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.4/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.4/Add",
+   "type": "/encoder/stages.2/stages.2.4/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.5/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.5/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.5/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.5/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.5/net/net.2/ReduceMean",
+   "type": "/encoder/stages.2/stages.2.5/net/net.2/ReduceMean",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.5/net/net.2/f/f.0/Conv",
+   "type": "/encoder/stages.2/stages.2.5/net/net.2/f/f.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.5/net/net.2/f/f.2/Conv",
+   "type": "/encoder/stages.2/stages.2.5/net/net.2/f/f.2/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.5/net/net.2/f/f.3/HardSigmoid",
+   "type": "/encoder/stages.2/stages.2.5/net/net.2/f/f.3/HardSigmoid",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.5/net/net.2/Mul",
+   "type": "/encoder/stages.2/stages.2.5/net/net.2/Mul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.5/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.5/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.5/Add",
+   "type": "/encoder/stages.2/stages.2.5/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.6/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.6/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.6/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.6/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.6/net/net.2/ReduceMean",
+   "type": "/encoder/stages.2/stages.2.6/net/net.2/ReduceMean",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.6/net/net.2/f/f.0/Conv",
+   "type": "/encoder/stages.2/stages.2.6/net/net.2/f/f.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.6/net/net.2/f/f.2/Conv",
+   "type": "/encoder/stages.2/stages.2.6/net/net.2/f/f.2/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.6/net/net.2/f/f.3/HardSigmoid",
+   "type": "/encoder/stages.2/stages.2.6/net/net.2/f/f.3/HardSigmoid",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.6/net/net.2/Mul",
+   "type": "/encoder/stages.2/stages.2.6/net/net.2/Mul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.6/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.6/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.6/Add",
+   "type": "/encoder/stages.2/stages.2.6/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.7/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.7/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.7/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.7/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.7/net/net.2/ReduceMean",
+   "type": "/encoder/stages.2/stages.2.7/net/net.2/ReduceMean",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.7/net/net.2/f/f.0/Conv",
+   "type": "/encoder/stages.2/stages.2.7/net/net.2/f/f.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.7/net/net.2/f/f.2/Conv",
+   "type": "/encoder/stages.2/stages.2.7/net/net.2/f/f.2/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.7/net/net.2/f/f.3/HardSigmoid",
+   "type": "/encoder/stages.2/stages.2.7/net/net.2/f/f.3/HardSigmoid",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.7/net/net.2/Mul",
+   "type": "/encoder/stages.2/stages.2.7/net/net.2/Mul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.7/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.7/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.7/Add",
+   "type": "/encoder/stages.2/stages.2.7/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat.2/lat.2.0/Conv",
+   "type": "/decoder/lat.2/lat.2.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.3/stages.3.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.3/stages.3.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.3/stages.3.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.3/stages.3.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.3/stages.3.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.2/ReduceMean",
+   "type": "/encoder/stages.3/stages.3.1/net/net.2/ReduceMean",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.2/f/f.0/Conv",
+   "type": "/encoder/stages.3/stages.3.1/net/net.2/f/f.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.2/f/f.2/Conv",
+   "type": "/encoder/stages.3/stages.3.1/net/net.2/f/f.2/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.2/f/f.3/HardSigmoid",
+   "type": "/encoder/stages.3/stages.3.1/net/net.2/f/f.3/HardSigmoid",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.2/Mul",
+   "type": "/encoder/stages.3/stages.3.1/net/net.2/Mul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.3/stages.3.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/Add",
+   "type": "/encoder/stages.3/stages.3.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.3/stages.3.2/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.3/stages.3.2/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.2/ReduceMean",
+   "type": "/encoder/stages.3/stages.3.2/net/net.2/ReduceMean",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.2/f/f.0/Conv",
+   "type": "/encoder/stages.3/stages.3.2/net/net.2/f/f.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.2/f/f.2/Conv",
+   "type": "/encoder/stages.3/stages.3.2/net/net.2/f/f.2/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.2/f/f.3/HardSigmoid",
+   "type": "/encoder/stages.3/stages.3.2/net/net.2/f/f.3/HardSigmoid",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.2/Mul",
+   "type": "/encoder/stages.3/stages.3.2/net/net.2/Mul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.3/stages.3.2/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/Add",
+   "type": "/encoder/stages.3/stages.3.2/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.3/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.3/stages.3.3/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.3/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.3/stages.3.3/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.3/net/net.2/ReduceMean",
+   "type": "/encoder/stages.3/stages.3.3/net/net.2/ReduceMean",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.3/net/net.2/f/f.0/Conv",
+   "type": "/encoder/stages.3/stages.3.3/net/net.2/f/f.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.3/net/net.2/f/f.2/Conv",
+   "type": "/encoder/stages.3/stages.3.3/net/net.2/f/f.2/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.3/net/net.2/f/f.3/HardSigmoid",
+   "type": "/encoder/stages.3/stages.3.3/net/net.2/f/f.3/HardSigmoid",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.3/net/net.2/Mul",
+   "type": "/encoder/stages.3/stages.3.3/net/net.2/Mul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.3/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.3/stages.3.3/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.3/Add",
+   "type": "/encoder/stages.3/stages.3.3/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.3/Add_output_0_DequantizeLinear_Output_converted_QNN_DATATYPE_FLOAT_16",
+   "type": "/encoder/stages.3/stages.3.3/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/n1/BatchNormalization",
+   "type": "/encoder/stages.3/stages.3.4/n1/BatchNormalization",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/n1/BatchNormalization_output_0_DequantizeLinear_Output_converted_QNN_DATATYPE_UFIXED_POINT_8",
+   "type": "/encoder/stages.3/stages.3.4/n1/BatchNormalization",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/k/Conv",
+   "type": "/encoder/stages.3/stages.3.4/k/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/q/Conv",
+   "type": "/encoder/stages.3/stages.3.4/q/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/v/Conv",
+   "type": "/encoder/stages.3/stages.3.4/v/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/k/Conv_output_0_DequantizeLinear_Output_0123",
+   "type": "/encoder/stages.3/stages.3.4/k/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/Reshape_1",
+   "type": "/encoder/stages.3/stages.3.4/Reshape",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/q/Conv_output_0_DequantizeLinear_Output_0123",
+   "type": "/encoder/stages.3/stages.3.4/q/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/Reshape",
+   "type": "/encoder/stages.3/stages.3.4/Reshape",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/v/Conv_output_0_DequantizeLinear_Output_0123",
+   "type": "/encoder/stages.3/stages.3.4/v/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/Reshape_2",
+   "type": "/encoder/stages.3/stages.3.4/Reshape",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/Transpose",
+   "type": "/encoder/stages.3/stages.3.4/Transpose",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/Transpose_2",
+   "type": "/encoder/stages.3/stages.3.4/Transpose",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/MatMul",
+   "type": "/encoder/stages.3/stages.3.4/MatMul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "node_Mul_1",
+   "type": "node",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/Softmax",
+   "type": "/encoder/stages.3/stages.3.4/Softmax",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/MatMul_1",
+   "type": "/encoder/stages.3/stages.3.4/MatMul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/Transpose_4",
+   "type": "/encoder/stages.3/stages.3.4/Transpose",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/Reshape_3",
+   "type": "/encoder/stages.3/stages.3.4/Reshape",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/Reshape_3_output_0_DequantizeLinear_Output_0231",
+   "type": "/encoder/stages.3/stages.3.4/Reshape",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/o/Conv",
+   "type": "/encoder/stages.3/stages.3.4/o/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/Add",
+   "type": "/encoder/stages.3/stages.3.4/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/Add_output_0_DequantizeLinear_Output_converted_QNN_DATATYPE_FLOAT_16",
+   "type": "/encoder/stages.3/stages.3.4/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/mlp/mlp.0/BatchNormalization",
+   "type": "/encoder/stages.3/stages.3.4/mlp/mlp.0/BatchNormalization",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/mlp/mlp.0/BatchNormalization_output_0_DequantizeLinear_Output_converted_QNN_DATATYPE_UFIXED_POINT_8",
+   "type": "/encoder/stages.3/stages.3.4/mlp/mlp.0/BatchNormalization",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/mlp/mlp.1/Conv",
+   "type": "/encoder/stages.3/stages.3.4/mlp/mlp.1/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/mlp/mlp.3/Conv",
+   "type": "/encoder/stages.3/stages.3.4/mlp/mlp.3/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/Add_1",
+   "type": "/encoder/stages.3/stages.3.4/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.4/Add_1_output_0_DequantizeLinear_Output_converted_QNN_DATATYPE_FLOAT_16",
+   "type": "/encoder/stages.3/stages.3.4/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/n1/BatchNormalization",
+   "type": "/encoder/stages.3/stages.3.5/n1/BatchNormalization",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/n1/BatchNormalization_output_0_DequantizeLinear_Output_converted_QNN_DATATYPE_UFIXED_POINT_8",
+   "type": "/encoder/stages.3/stages.3.5/n1/BatchNormalization",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/k/Conv",
+   "type": "/encoder/stages.3/stages.3.5/k/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/q/Conv",
+   "type": "/encoder/stages.3/stages.3.5/q/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/v/Conv",
+   "type": "/encoder/stages.3/stages.3.5/v/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/k/Conv_output_0_DequantizeLinear_Output_0123",
+   "type": "/encoder/stages.3/stages.3.5/k/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/Reshape_1",
+   "type": "/encoder/stages.3/stages.3.5/Reshape",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/q/Conv_output_0_DequantizeLinear_Output_0123",
+   "type": "/encoder/stages.3/stages.3.5/q/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/Reshape",
+   "type": "/encoder/stages.3/stages.3.5/Reshape",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/v/Conv_output_0_DequantizeLinear_Output_0123",
+   "type": "/encoder/stages.3/stages.3.5/v/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/Reshape_2",
+   "type": "/encoder/stages.3/stages.3.5/Reshape",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/Transpose",
+   "type": "/encoder/stages.3/stages.3.5/Transpose",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/Transpose_2",
+   "type": "/encoder/stages.3/stages.3.5/Transpose",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/MatMul",
+   "type": "/encoder/stages.3/stages.3.5/MatMul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "node_Mul_3",
+   "type": "node",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/Softmax",
+   "type": "/encoder/stages.3/stages.3.5/Softmax",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/MatMul_1",
+   "type": "/encoder/stages.3/stages.3.5/MatMul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/Transpose_4",
+   "type": "/encoder/stages.3/stages.3.5/Transpose",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/Reshape_3",
+   "type": "/encoder/stages.3/stages.3.5/Reshape",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/Reshape_3_output_0_DequantizeLinear_Output_0231",
+   "type": "/encoder/stages.3/stages.3.5/Reshape",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/o/Conv",
+   "type": "/encoder/stages.3/stages.3.5/o/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/Add",
+   "type": "/encoder/stages.3/stages.3.5/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/Add_output_0_DequantizeLinear_Output_converted_QNN_DATATYPE_FLOAT_16",
+   "type": "/encoder/stages.3/stages.3.5/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/mlp/mlp.0/BatchNormalization",
+   "type": "/encoder/stages.3/stages.3.5/mlp/mlp.0/BatchNormalization",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/mlp/mlp.0/BatchNormalization_output_0_DequantizeLinear_Output_converted_QNN_DATATYPE_UFIXED_POINT_8",
+   "type": "/encoder/stages.3/stages.3.5/mlp/mlp.0/BatchNormalization",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/mlp/mlp.1/Conv",
+   "type": "/encoder/stages.3/stages.3.5/mlp/mlp.1/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/mlp/mlp.3/Conv",
+   "type": "/encoder/stages.3/stages.3.5/mlp/mlp.3/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.5/Add_1",
+   "type": "/encoder/stages.3/stages.3.5/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/neck.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/neck.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/neck.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.0/Add",
+   "type": "/encoder/neck.0/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/neck.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/neck.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/neck.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.1/Add",
+   "type": "/encoder/neck.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.2/net/net.0/net.0.0/Conv",
+   "type": "/encoder/neck.2/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.2/net/net.1/net.1.0/Conv",
+   "type": "/encoder/neck.2/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.2/net/net.3/net.3.0/Conv",
+   "type": "/encoder/neck.2/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.2/Add",
+   "type": "/encoder/neck.2/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat.3/lat.3.0/Conv",
+   "type": "/decoder/lat.3/lat.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Resize",
+   "type": "/decoder/Resize",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Add",
+   "type": "/decoder/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.2/net/net.0/net.0.0/Conv",
+   "type": "/decoder/ref.2/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.2/net/net.1/net.1.0/Conv",
+   "type": "/decoder/ref.2/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.2/net/net.3/net.3.0/Conv",
+   "type": "/decoder/ref.2/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.2/Add",
+   "type": "/decoder/ref.2/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Resize_1",
+   "type": "/decoder/Resize",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Add_1",
+   "type": "/decoder/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.1/net/net.0/net.0.0/Conv",
+   "type": "/decoder/ref.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.1/net/net.1/net.1.0/Conv",
+   "type": "/decoder/ref.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.1/net/net.3/net.3.0/Conv",
+   "type": "/decoder/ref.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.1/Add",
+   "type": "/decoder/ref.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Resize_2",
+   "type": "/decoder/Resize",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Add_2",
+   "type": "/decoder/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.0/net/net.0/net.0.0/Conv",
+   "type": "/decoder/ref.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.0/net/net.1/net.1.0/Conv",
+   "type": "/decoder/ref.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.0/net/net.3/net.3.0/Conv",
+   "type": "/decoder/ref.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.0/Add",
+   "type": "/decoder/ref.0/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Resize_3",
+   "type": "/decoder/Resize",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Add_3",
+   "type": "/decoder/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref0/net/net.0/net.0.0/Conv",
+   "type": "/decoder/ref0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref0/net/net.1/net.1.0/Conv",
+   "type": "/decoder/ref0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref0/net/net.3/net.3.0/Conv",
+   "type": "/decoder/ref0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref0/Add",
+   "type": "/decoder/ref0/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref0/Add_output_0_DequantizeLinear_Output_converted_QNN_DATATYPE_FLOAT_16",
+   "type": "/decoder/ref0/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/head/head.1/Conv",
+   "type": "/head/head.1/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "output_0_0231",
+   "type": "output",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "QNN_DATATYPE_FLOAT_16_converted_output_0_QNN_DATATYPE_FLOAT_32",
+   "type": "QNN",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "Output",
+   "type": "Output",
+   "compute_unit": "NPU"
+  }
+ ]
+}

models.py

@@ -0,0 +1,149 @@
+"""Blue Marble model architecture (Apache-2.0)."""
+import torch, torch.nn as nn, torch.nn.functional as F
+S2_BANDS = ['B01', 'B02', 'B03', 'B04', 'B05', 'B06', 'B07', 'B08', 'B8A', 'B09', 'B10', 'B11', 'B12']
+
+def conv_bn(i, o, k=3, s=1, g=1, d=1, act=True):
+    layers = [nn.Conv2d(i, o, k, s, d * (k // 2), dilation=d, groups=g, bias=False), nn.BatchNorm2d(o)]
+    if act:
+        layers.append(nn.ReLU6(inplace=True))
+    return nn.Sequential(*layers)
+
+class SE(nn.Module):
+
+    def __init__(self, c, r=4):
+        super().__init__()
+        h = max(8, c // r)
+        self.f = nn.Sequential(nn.Conv2d(c, h, 1), nn.ReLU6(inplace=True), nn.Conv2d(h, c, 1), nn.Hardsigmoid())
+
+    def forward(self, x):
+        return x * self.f(x.mean((2, 3), keepdim=True))
+
+class IRB(nn.Module):
+
+    def __init__(self, i, o, s=1, e=3, k=5, se=False, d=1):
+        super().__init__()
+        h = int(round(i * e))
+        self.res = s == 1 and i == o
+        self.net = nn.Sequential(conv_bn(i, h, 1) if e != 1 else nn.Identity(), conv_bn(h, h, k, s, g=h, d=d), SE(h) if se else nn.Identity(), conv_bn(h, o, 1, act=False))
+
+    def forward(self, x):
+        y = self.net(x)
+        return x + y if self.res else y
+
+class SpectralStem(nn.Module):
+
+    def __init__(self, c, in_bands=13):
+        super().__init__()
+        self.spectral = nn.Conv2d(in_bands, c, 1, bias=False)
+        self.bn = nn.BatchNorm2d(c)
+        self.act = nn.ReLU6(inplace=True)
+        self.spatial = conv_bn(c, c, 3, g=c)
+        self.fuse = conv_bn(c, c, 1)
+
+    def forward(self, x):
+        return self.fuse(self.spatial(self.act(self.bn(self.spectral(x)))))
+
+    @torch.no_grad()
+    def slice_bands(self, idx):
+        new = nn.Conv2d(len(idx), self.spectral.out_channels, 1, bias=False).to(self.spectral.weight.device)
+        new.weight.copy_(self.spectral.weight[:, idx])
+        self.spectral = new
+
+class Attn2d(nn.Module):
+
+    def __init__(self, c, heads=4, mlp=2):
+        super().__init__()
+        self.h, self.d = (heads, c // heads)
+        self.n1, self.q, self.k, self.v, self.o = (nn.BatchNorm2d(c), nn.Conv2d(c, c, 1), nn.Conv2d(c, c, 1), nn.Conv2d(c, c, 1), nn.Conv2d(c, c, 1))
+        self.mlp = nn.Sequential(nn.BatchNorm2d(c), nn.Conv2d(c, c * mlp, 1), nn.ReLU6(inplace=True), nn.Conv2d(c * mlp, c, 1))
+
+    def forward(self, x):
+        b, c, h, w = x.shape
+        n = self.n1(x)
+        sp = lambda t: t.reshape(b, self.h, self.d, h * w).transpose(2, 3)
+        q, k, v = (sp(self.q(n)), sp(self.k(n)), sp(self.v(n)))
+        a = (q.float() @ k.float().transpose(2, 3) / self.d ** 0.5).softmax(-1)
+        y = (a @ v.float()).to(x.dtype).transpose(2, 3).reshape(b, c, h, w)
+        x = x + self.o(y)
+        return x + self.mlp(x)
+ARCHS = {'pico': dict(stem=8, w=(8, 16, 24, 40), d=(1, 2, 2, 2), e=2, k=3, rates=(1,), dec=16, full=False, se=False), 'nano': dict(stem=12, w=(16, 24, 48, 80), d=(1, 2, 3, 2), e=3, k=5, rates=(1, 2), dec=24, full=True, se=False), 'tiny': dict(stem=16, w=(24, 40, 80, 160), d=(2, 2, 4, 3), e=3, k=5, rates=(1, 2, 4), dec=40, full=True, se=True), 'small': dict(stem=24, w=(32, 56, 112, 208), d=(2, 3, 5, 3), e=3, k=7, rates=(1, 2, 4), dec=48, full=True, se=True), 'medium': dict(stem=32, w=(48, 96, 192, 320), d=(2, 3, 8, 4), e=4, k=7, rates=(1, 2, 4), dec=96, full=True, se=True, attn=2)}
+
+class Encoder(nn.Module):
+
+    def __init__(self, cfg, in_bands=13):
+        super().__init__()
+        self.stem = SpectralStem(cfg['stem'], in_bands)
+        prev, self.stages = (cfg['stem'], nn.ModuleList())
+        for si, (w, n) in enumerate(zip(cfg['w'], cfg['d'])):
+            blocks = [IRB(prev, w, 2, cfg['e'], 3)]
+            blocks += [IRB(w, w, 1, cfg['e'], cfg['k'], se=cfg['se'] and si >= 2) for _ in range(n - 1)]
+            if si == len(cfg['w']) - 1:
+                blocks += [Attn2d(w, heads=8) for _ in range(cfg.get('attn', 0))]
+            self.stages.append(nn.Sequential(*blocks))
+            prev = w
+        self.neck = nn.ModuleList((IRB(prev, prev, 1, 2, 3, d=r) for r in cfg['rates']))
+        self.channels = (cfg['stem'], *cfg['w'])
+
+    def forward(self, x):
+        out = [self.stem(x)]
+        for s in self.stages:
+            out.append(s(out[-1]))
+        y = out[-1]
+        for n in self.neck:
+            y = n(y)
+        out[-1] = y
+        return out
+
+class FPNDecoder(nn.Module):
+
+    def __init__(self, chs, dec, full=True):
+        super().__init__()
+        self.full = full
+        self.lat = nn.ModuleList((conv_bn(c, dec, 1) for c in chs[1:]))
+        self.ref = nn.ModuleList((IRB(dec, dec, 1, 2, 3) for _ in chs[1:-1]))
+        if full:
+            self.lat0 = conv_bn(chs[0], dec, 1)
+            self.ref0 = IRB(dec, dec, 1, 2, 3)
+        self.out_channels = dec
+
+    def forward(self, f):
+        y = self.lat[-1](f[-1])
+        for i in range(len(self.ref) - 1, -1, -1):
+            y = F.interpolate(y, scale_factor=2.0, mode='bilinear', align_corners=False) + self.lat[i](f[i + 1])
+            y = self.ref[i](y)
+        if self.full:
+            y = F.interpolate(y, scale_factor=2.0, mode='bilinear', align_corners=False) + self.lat0(f[0])
+            y = self.ref0(y)
+        return y
+
+class BlueMarble(nn.Module):
+
+    def __init__(self, arch='tiny', num_classes=2, in_bands=13):
+        super().__init__()
+        cfg = ARCHS[arch]
+        self.arch, self.upsample = (arch, 1 if cfg['full'] else 2)
+        self.encoder = Encoder(cfg, in_bands)
+        self.decoder = FPNDecoder(self.encoder.channels, cfg['dec'], cfg['full'])
+        self.head = nn.Sequential(nn.Dropout2d(0.1), nn.Conv2d(cfg['dec'], num_classes, 1))
+
+    def forward(self, x):
+        y = self.head(self.decoder(self.encoder(x)))
+        if self.upsample > 1:
+            y = F.interpolate(y, scale_factor=float(self.upsample), mode='bilinear', align_corners=False)
+        return y
+
+def fast_layout(model, in_shape, device='cuda'):
+    import torch.nn.functional as F
+    if not str(device).startswith('cuda'):
+        return (model, False)
+    try:
+        m = model.to(memory_format=torch.channels_last)
+        x = torch.randn(in_shape, device=device).to(memory_format=torch.channels_last)
+        with torch.autocast('cuda', dtype=torch.bfloat16):
+            out = m(x)
+        out.float().mean().backward()
+        m.zero_grad(set_to_none=True)
+        return (m, True)
+    except RuntimeError as e:
+        print('channels_last unavailable, using NCHW:', str(e)[:120], flush=True)
+        return (model.to(memory_format=torch.contiguous_format), False)

nano/config.json

@@ -0,0 +1 @@
+{"arch": "nano", "num_classes": 2}

nano/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2246cb75f5e5ad22f80eca4dbc840c14f5120f7c2d41f44c6bd526facac33dc8
+size 821152

nano/model_fp32.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8e18ebbd24e416ad5b4fd1b17b6e610792d4d53863f85065cbda5f2cf8625981
+size 789256

nano/model_int8.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2bef88ba00f8b616f642c54a809ffa8c85a3bbbccd66b5f14b2807b194cdb5b3
+size 359170

nano/model_qcs8275.dlc

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:89ae8e4576aecacbcd4fba59eb9806258f2b5a6d6631041771b6905e0e97912c
+size 526808

nano/ventuno_q_profile.json

@@ -0,0 +1,416 @@
+{
+ "device": "Arduino VENTUNO Q (Qualcomm Dragonwing QCS8275)",
+ "device_attributes": [
+  "os:ubuntu",
+  "chipset:qualcomm-qcs8275",
+  "chipset:qcs8275",
+  "chipset:qualcomm-dragonwing-iq-8275",
+  "hexagon:v75",
+  "soc-model:82",
+  "framework:qnn"
+ ],
+ "source": "Qualcomm AI Hub on-device profile job",
+ "profiled_at_utc": "2026-11-06 13:09:58",
+ "file": "model_qcs8275.dlc",
+ "precision": "int8",
+ "input": [
+  1,
+  13,
+  256,
+  256
+ ],
+ "ops_total": 76,
+ "ops_per_compute_unit": {
+  "NPU": 76,
+  "GPU": 0,
+  "CPU": 0
+ },
+ "inference_time_us": 975,
+ "inference_peak_memory_bytes": 12652544,
+ "first_load_time_us": 1223189,
+ "warm_load_time_us": 186604,
+ "note": "peak memory is runtime working memory (activations and buffers), not the weight file size",
+ "on_device_vs_host_agreement": null,
+ "ops": [
+  {
+   "name": "Input",
+   "type": "Input",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "QNN_DATATYPE_FLOAT_32_converted_image_QNN_DATATYPE_UFIXED_POINT_8",
+   "type": "QNN",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "image_0231",
+   "type": "image",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stem/spectral/Conv",
+   "type": "/encoder/stem/spectral/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stem/spatial/spatial.0/Conv",
+   "type": "/encoder/stem/spatial/spatial.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stem/fuse/fuse.0/Conv",
+   "type": "/encoder/stem/fuse/fuse.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat0/lat0.0/Conv",
+   "type": "/decoder/lat0/lat0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.0/stages.0.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.0/stages.0.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.0/stages.0.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat.0/lat.0.0/Conv",
+   "type": "/decoder/lat.0/lat.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.1/stages.1.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.1/stages.1.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.1/stages.1.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.1/stages.1.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.1/stages.1.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.1/stages.1.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.1/Add",
+   "type": "/encoder/stages.1/stages.1.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat.1/lat.1.0/Conv",
+   "type": "/decoder/lat.1/lat.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/Add",
+   "type": "/encoder/stages.2/stages.2.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.2/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.2/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.2/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/Add",
+   "type": "/encoder/stages.2/stages.2.2/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat.2/lat.2.0/Conv",
+   "type": "/decoder/lat.2/lat.2.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.3/stages.3.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.3/stages.3.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.3/stages.3.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.3/stages.3.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.3/stages.3.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.3/stages.3.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/Add",
+   "type": "/encoder/stages.3/stages.3.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/neck.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/neck.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/neck.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.0/Add",
+   "type": "/encoder/neck.0/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/neck.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/neck.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/neck.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.1/Add",
+   "type": "/encoder/neck.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat.3/lat.3.0/Conv",
+   "type": "/decoder/lat.3/lat.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Resize",
+   "type": "/decoder/Resize",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Add",
+   "type": "/decoder/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.2/net/net.0/net.0.0/Conv",
+   "type": "/decoder/ref.2/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.2/net/net.1/net.1.0/Conv",
+   "type": "/decoder/ref.2/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.2/net/net.3/net.3.0/Conv",
+   "type": "/decoder/ref.2/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.2/Add",
+   "type": "/decoder/ref.2/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Resize_1",
+   "type": "/decoder/Resize",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Add_1",
+   "type": "/decoder/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.1/net/net.0/net.0.0/Conv",
+   "type": "/decoder/ref.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.1/net/net.1/net.1.0/Conv",
+   "type": "/decoder/ref.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.1/net/net.3/net.3.0/Conv",
+   "type": "/decoder/ref.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.1/Add",
+   "type": "/decoder/ref.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Resize_2",
+   "type": "/decoder/Resize",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Add_2",
+   "type": "/decoder/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.0/net/net.0/net.0.0/Conv",
+   "type": "/decoder/ref.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.0/net/net.1/net.1.0/Conv",
+   "type": "/decoder/ref.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.0/net/net.3/net.3.0/Conv",
+   "type": "/decoder/ref.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.0/Add",
+   "type": "/decoder/ref.0/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Resize_3",
+   "type": "/decoder/Resize",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Add_3",
+   "type": "/decoder/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref0/net/net.0/net.0.0/Conv",
+   "type": "/decoder/ref0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref0/net/net.1/net.1.0/Conv",
+   "type": "/decoder/ref0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref0/net/net.3/net.3.0/Conv",
+   "type": "/decoder/ref0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref0/Add",
+   "type": "/decoder/ref0/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref0/Add_output_0_DequantizeLinear_Output_converted_QNN_DATATYPE_FLOAT_16",
+   "type": "/decoder/ref0/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/head/head.1/Conv",
+   "type": "/head/head.1/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "output_0_0231",
+   "type": "output",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "QNN_DATATYPE_FLOAT_16_converted_output_0_QNN_DATATYPE_FLOAT_32",
+   "type": "QNN",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "Output",
+   "type": "Output",
+   "compute_unit": "NPU"
+  }
+ ]
+}

pico/config.json

@@ -0,0 +1 @@
+{"arch": "pico", "num_classes": 2}

pico/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4eb5b2f285c96e5bcc396a5f137f559eaebf0475f76a6ba854dbd73fa3b66124
+size 162680

pico/model_fp32.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e759ec5a58042c6b9c48a6db552dc47215be57e46f800ba3986d7cdbf2f75c6
+size 157495

pico/model_int8.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:edd2682e4e7c7a711c270de19e103ae2181390f61b0f213b40b49f1600482227
+size 141995

pico/model_qcs8275.dlc

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5ed69ecc15780bfd0e35f18b738b5a1ab9437ef07705562400714300231e28ee
+size 208424

pico/ventuno_q_profile.json

@@ -0,0 +1,346 @@
+{
+ "device": "Arduino VENTUNO Q (Qualcomm Dragonwing QCS8275)",
+ "device_attributes": [
+  "os:ubuntu",
+  "chipset:qualcomm-qcs8275",
+  "chipset:qcs8275",
+  "chipset:qualcomm-dragonwing-iq-8275",
+  "hexagon:v75",
+  "soc-model:82",
+  "framework:qnn"
+ ],
+ "source": "Qualcomm AI Hub on-device profile job",
+ "profiled_at_utc": "2026-11-06 14:12:32",
+ "file": "model_qcs8275.dlc",
+ "precision": "int8",
+ "input": [
+  1,
+  13,
+  256,
+  256
+ ],
+ "ops_total": 62,
+ "ops_per_compute_unit": {
+  "NPU": 62,
+  "GPU": 0,
+  "CPU": 0
+ },
+ "inference_time_us": 977,
+ "inference_peak_memory_bytes": 12648448,
+ "first_load_time_us": 892803,
+ "warm_load_time_us": 184102,
+ "note": "peak memory is runtime working memory (activations and buffers), not the weight file size",
+ "on_device_vs_host_agreement": null,
+ "ops": [
+  {
+   "name": "Input",
+   "type": "Input",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "QNN_DATATYPE_FLOAT_32_converted_image_QNN_DATATYPE_UFIXED_POINT_8",
+   "type": "QNN",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "image_0231",
+   "type": "image",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stem/spectral/Conv",
+   "type": "/encoder/stem/spectral/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stem/spatial/spatial.0/Conv",
+   "type": "/encoder/stem/spatial/spatial.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stem/fuse/fuse.0/Conv",
+   "type": "/encoder/stem/fuse/fuse.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.0/stages.0.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.0/stages.0.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.0/stages.0.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat.0/lat.0.0/Conv",
+   "type": "/decoder/lat.0/lat.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.1/stages.1.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.1/stages.1.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.1/stages.1.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.1/stages.1.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.1/stages.1.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.1/stages.1.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.1/Add",
+   "type": "/encoder/stages.1/stages.1.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat.1/lat.1.0/Conv",
+   "type": "/decoder/lat.1/lat.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/Add",
+   "type": "/encoder/stages.2/stages.2.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat.2/lat.2.0/Conv",
+   "type": "/decoder/lat.2/lat.2.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.3/stages.3.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.3/stages.3.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.3/stages.3.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.3/stages.3.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.3/stages.3.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.3/stages.3.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/Add",
+   "type": "/encoder/stages.3/stages.3.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/neck.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/neck.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/neck.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.0/Add",
+   "type": "/encoder/neck.0/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat.3/lat.3.0/Conv",
+   "type": "/decoder/lat.3/lat.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Resize",
+   "type": "/decoder/Resize",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Add",
+   "type": "/decoder/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.2/net/net.0/net.0.0/Conv",
+   "type": "/decoder/ref.2/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.2/net/net.1/net.1.0/Conv",
+   "type": "/decoder/ref.2/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.2/net/net.3/net.3.0/Conv",
+   "type": "/decoder/ref.2/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.2/Add",
+   "type": "/decoder/ref.2/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Resize_1",
+   "type": "/decoder/Resize",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Add_1",
+   "type": "/decoder/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.1/net/net.0/net.0.0/Conv",
+   "type": "/decoder/ref.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.1/net/net.1/net.1.0/Conv",
+   "type": "/decoder/ref.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.1/net/net.3/net.3.0/Conv",
+   "type": "/decoder/ref.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.1/Add",
+   "type": "/decoder/ref.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Resize_2",
+   "type": "/decoder/Resize",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Add_2",
+   "type": "/decoder/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.0/net/net.0/net.0.0/Conv",
+   "type": "/decoder/ref.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.0/net/net.1/net.1.0/Conv",
+   "type": "/decoder/ref.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.0/net/net.3/net.3.0/Conv",
+   "type": "/decoder/ref.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.0/Add",
+   "type": "/decoder/ref.0/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.0/Add_output_0_DequantizeLinear_Output_converted_QNN_DATATYPE_FLOAT_16",
+   "type": "/decoder/ref.0/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/head/head.1/Conv",
+   "type": "/head/head.1/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/Resize",
+   "type": "/Resize",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "output_0_0231",
+   "type": "output",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "QNN_DATATYPE_FLOAT_16_converted_output_0_QNN_DATATYPE_FLOAT_32",
+   "type": "QNN",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "Output",
+   "type": "Output",
+   "compute_unit": "NPU"
+  }
+ ]
+}

small/config.json

@@ -0,0 +1 @@
+{"arch": "small", "num_classes": 2}

small/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb653a9af6e0ed6d51e7abaee9402273a0d96e34bcc2dc5ba9843e10331f9dae
+size 9764600

small/model_fp32.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aa5467df5ff60cea847d06d3ecbc5e6559fd3a972d454113da6fed81853eaa6b
+size 9631708

small/model_int8.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:41b1731e04cb764e33c44d83e030ce4dcf86e8193cc8e306e006be1cae0e1f64
+size 2863586

small/model_qcs8275.dlc

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c01fcbea8f5387204a63c74a67413dc2b5939febec282a3a5542ff72b25cbcc
+size 3572184

small/ventuno_q_profile.json

@@ -0,0 +1,686 @@
+{
+ "device": "Arduino VENTUNO Q (Qualcomm Dragonwing QCS8275)",
+ "device_attributes": [
+  "os:ubuntu",
+  "chipset:qualcomm-qcs8275",
+  "chipset:qcs8275",
+  "chipset:qualcomm-dragonwing-iq-8275",
+  "hexagon:v75",
+  "soc-model:82",
+  "framework:qnn"
+ ],
+ "source": "Qualcomm AI Hub on-device profile job",
+ "profiled_at_utc": "2026-11-06 14:12:46",
+ "file": "model_qcs8275.dlc",
+ "precision": "int8",
+ "input": [
+  1,
+  13,
+  256,
+  256
+ ],
+ "ops_total": 130,
+ "ops_per_compute_unit": {
+  "NPU": 130,
+  "GPU": 0,
+  "CPU": 0
+ },
+ "inference_time_us": 1503,
+ "inference_peak_memory_bytes": 12660736,
+ "first_load_time_us": 1952166,
+ "warm_load_time_us": 169220,
+ "note": "peak memory is runtime working memory (activations and buffers), not the weight file size",
+ "on_device_vs_host_agreement": null,
+ "ops": [
+  {
+   "name": "Input",
+   "type": "Input",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "QNN_DATATYPE_FLOAT_32_converted_image_QNN_DATATYPE_UFIXED_POINT_8",
+   "type": "QNN",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "image_0231",
+   "type": "image",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stem/spectral/Conv",
+   "type": "/encoder/stem/spectral/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stem/spatial/spatial.0/Conv",
+   "type": "/encoder/stem/spatial/spatial.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stem/fuse/fuse.0/Conv",
+   "type": "/encoder/stem/fuse/fuse.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat0/lat0.0/Conv",
+   "type": "/decoder/lat0/lat0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.0/stages.0.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.0/stages.0.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.0/stages.0.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.0/stages.0.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.0/stages.0.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.0/stages.0.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.1/Add",
+   "type": "/encoder/stages.0/stages.0.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat.0/lat.0.0/Conv",
+   "type": "/decoder/lat.0/lat.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.1/stages.1.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.1/stages.1.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.1/stages.1.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.1/stages.1.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.1/stages.1.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.1/stages.1.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.1/Add",
+   "type": "/encoder/stages.1/stages.1.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.2/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.1/stages.1.2/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.2/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.1/stages.1.2/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.2/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.1/stages.1.2/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.2/Add",
+   "type": "/encoder/stages.1/stages.1.2/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat.1/lat.1.0/Conv",
+   "type": "/decoder/lat.1/lat.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.2/ReduceMean",
+   "type": "/encoder/stages.2/stages.2.1/net/net.2/ReduceMean",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.2/f/f.0/Conv",
+   "type": "/encoder/stages.2/stages.2.1/net/net.2/f/f.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.2/f/f.2/Conv",
+   "type": "/encoder/stages.2/stages.2.1/net/net.2/f/f.2/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.2/f/f.3/HardSigmoid",
+   "type": "/encoder/stages.2/stages.2.1/net/net.2/f/f.3/HardSigmoid",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.2/Mul",
+   "type": "/encoder/stages.2/stages.2.1/net/net.2/Mul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/Add",
+   "type": "/encoder/stages.2/stages.2.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.2/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.2/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.2/ReduceMean",
+   "type": "/encoder/stages.2/stages.2.2/net/net.2/ReduceMean",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.2/f/f.0/Conv",
+   "type": "/encoder/stages.2/stages.2.2/net/net.2/f/f.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.2/f/f.2/Conv",
+   "type": "/encoder/stages.2/stages.2.2/net/net.2/f/f.2/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.2/f/f.3/HardSigmoid",
+   "type": "/encoder/stages.2/stages.2.2/net/net.2/f/f.3/HardSigmoid",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.2/Mul",
+   "type": "/encoder/stages.2/stages.2.2/net/net.2/Mul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.2/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/Add",
+   "type": "/encoder/stages.2/stages.2.2/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.3/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.3/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.2/ReduceMean",
+   "type": "/encoder/stages.2/stages.2.3/net/net.2/ReduceMean",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.2/f/f.0/Conv",
+   "type": "/encoder/stages.2/stages.2.3/net/net.2/f/f.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.2/f/f.2/Conv",
+   "type": "/encoder/stages.2/stages.2.3/net/net.2/f/f.2/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.2/f/f.3/HardSigmoid",
+   "type": "/encoder/stages.2/stages.2.3/net/net.2/f/f.3/HardSigmoid",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.2/Mul",
+   "type": "/encoder/stages.2/stages.2.3/net/net.2/Mul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.3/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/Add",
+   "type": "/encoder/stages.2/stages.2.3/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.4/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.4/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.4/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.4/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.4/net/net.2/ReduceMean",
+   "type": "/encoder/stages.2/stages.2.4/net/net.2/ReduceMean",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.4/net/net.2/f/f.0/Conv",
+   "type": "/encoder/stages.2/stages.2.4/net/net.2/f/f.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.4/net/net.2/f/f.2/Conv",
+   "type": "/encoder/stages.2/stages.2.4/net/net.2/f/f.2/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.4/net/net.2/f/f.3/HardSigmoid",
+   "type": "/encoder/stages.2/stages.2.4/net/net.2/f/f.3/HardSigmoid",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.4/net/net.2/Mul",
+   "type": "/encoder/stages.2/stages.2.4/net/net.2/Mul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.4/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.4/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.4/Add",
+   "type": "/encoder/stages.2/stages.2.4/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat.2/lat.2.0/Conv",
+   "type": "/decoder/lat.2/lat.2.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.3/stages.3.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.3/stages.3.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.3/stages.3.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.3/stages.3.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.3/stages.3.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.2/ReduceMean",
+   "type": "/encoder/stages.3/stages.3.1/net/net.2/ReduceMean",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.2/f/f.0/Conv",
+   "type": "/encoder/stages.3/stages.3.1/net/net.2/f/f.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.2/f/f.2/Conv",
+   "type": "/encoder/stages.3/stages.3.1/net/net.2/f/f.2/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.2/f/f.3/HardSigmoid",
+   "type": "/encoder/stages.3/stages.3.1/net/net.2/f/f.3/HardSigmoid",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.2/Mul",
+   "type": "/encoder/stages.3/stages.3.1/net/net.2/Mul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.3/stages.3.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/Add",
+   "type": "/encoder/stages.3/stages.3.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.3/stages.3.2/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.3/stages.3.2/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.2/ReduceMean",
+   "type": "/encoder/stages.3/stages.3.2/net/net.2/ReduceMean",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.2/f/f.0/Conv",
+   "type": "/encoder/stages.3/stages.3.2/net/net.2/f/f.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.2/f/f.2/Conv",
+   "type": "/encoder/stages.3/stages.3.2/net/net.2/f/f.2/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.2/f/f.3/HardSigmoid",
+   "type": "/encoder/stages.3/stages.3.2/net/net.2/f/f.3/HardSigmoid",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.2/Mul",
+   "type": "/encoder/stages.3/stages.3.2/net/net.2/Mul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.3/stages.3.2/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/Add",
+   "type": "/encoder/stages.3/stages.3.2/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/neck.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/neck.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/neck.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.0/Add",
+   "type": "/encoder/neck.0/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/neck.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/neck.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/neck.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.1/Add",
+   "type": "/encoder/neck.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.2/net/net.0/net.0.0/Conv",
+   "type": "/encoder/neck.2/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.2/net/net.1/net.1.0/Conv",
+   "type": "/encoder/neck.2/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.2/net/net.3/net.3.0/Conv",
+   "type": "/encoder/neck.2/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.2/Add",
+   "type": "/encoder/neck.2/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat.3/lat.3.0/Conv",
+   "type": "/decoder/lat.3/lat.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Resize",
+   "type": "/decoder/Resize",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Add",
+   "type": "/decoder/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.2/net/net.0/net.0.0/Conv",
+   "type": "/decoder/ref.2/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.2/net/net.1/net.1.0/Conv",
+   "type": "/decoder/ref.2/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.2/net/net.3/net.3.0/Conv",
+   "type": "/decoder/ref.2/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.2/Add",
+   "type": "/decoder/ref.2/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Resize_1",
+   "type": "/decoder/Resize",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Add_1",
+   "type": "/decoder/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.1/net/net.0/net.0.0/Conv",
+   "type": "/decoder/ref.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.1/net/net.1/net.1.0/Conv",
+   "type": "/decoder/ref.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.1/net/net.3/net.3.0/Conv",
+   "type": "/decoder/ref.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.1/Add",
+   "type": "/decoder/ref.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Resize_2",
+   "type": "/decoder/Resize",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Add_2",
+   "type": "/decoder/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.0/net/net.0/net.0.0/Conv",
+   "type": "/decoder/ref.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.0/net/net.1/net.1.0/Conv",
+   "type": "/decoder/ref.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.0/net/net.3/net.3.0/Conv",
+   "type": "/decoder/ref.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.0/Add",
+   "type": "/decoder/ref.0/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Resize_3",
+   "type": "/decoder/Resize",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Add_3",
+   "type": "/decoder/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref0/net/net.0/net.0.0/Conv",
+   "type": "/decoder/ref0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref0/net/net.1/net.1.0/Conv",
+   "type": "/decoder/ref0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref0/net/net.3/net.3.0/Conv",
+   "type": "/decoder/ref0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref0/Add",
+   "type": "/decoder/ref0/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref0/Add_output_0_DequantizeLinear_Output_converted_QNN_DATATYPE_FLOAT_16",
+   "type": "/decoder/ref0/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/head/head.1/Conv",
+   "type": "/head/head.1/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "output_0_0231",
+   "type": "output",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "QNN_DATATYPE_FLOAT_16_converted_output_0_QNN_DATATYPE_FLOAT_32",
+   "type": "QNN",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "Output",
+   "type": "Output",
+   "compute_unit": "NPU"
+  }
+ ]
+}

tiny/config.json

@@ -0,0 +1 @@
+{"arch": "tiny", "num_classes": 2}

tiny/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ce9ca37599e54a551393e5e387b2a1f11fda1bfd774b6baf969d73c1f791b5ea
+size 5151672

tiny/model_fp32.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f259f2305e6c2a5982ea9199733666f476eb7401ff3ca180dc9fff76662ae10f
+size 5065044

tiny/model_int8.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d966fa4b5833a39c163f14882770e1ce56e092aa0833917a7c2dd2522a248070
+size 1616031

tiny/model_qcs8275.dlc

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16cba6d2a567409b5baabdc7a91e9284391d653308619668355ab3ecbca82c8b
+size 2100736

tiny/ventuno_q_profile.json

@@ -0,0 +1,621 @@
+{
+ "device": "Arduino VENTUNO Q (Qualcomm Dragonwing QCS8275)",
+ "device_attributes": [
+  "os:ubuntu",
+  "chipset:qualcomm-qcs8275",
+  "chipset:qcs8275",
+  "chipset:qualcomm-dragonwing-iq-8275",
+  "hexagon:v75",
+  "soc-model:82",
+  "framework:qnn"
+ ],
+ "source": "Qualcomm AI Hub on-device profile job",
+ "profiled_at_utc": "2026-11-06 14:51:49",
+ "file": "model_qcs8275.dlc",
+ "precision": "int8",
+ "input": [
+  1,
+  13,
+  256,
+  256
+ ],
+ "ops_total": 117,
+ "ops_per_compute_unit": {
+  "NPU": 117,
+  "GPU": 0,
+  "CPU": 0
+ },
+ "inference_time_us": 1272,
+ "inference_peak_memory_bytes": 12656640,
+ "first_load_time_us": 2069567,
+ "warm_load_time_us": 181232,
+ "note": "peak memory is runtime working memory (activations and buffers), not the weight file size",
+ "on_device_vs_host_agreement": null,
+ "ops": [
+  {
+   "name": "Input",
+   "type": "Input",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "QNN_DATATYPE_FLOAT_32_converted_image_QNN_DATATYPE_UFIXED_POINT_8",
+   "type": "QNN",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "image_0231",
+   "type": "image",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stem/spectral/Conv",
+   "type": "/encoder/stem/spectral/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stem/spatial/spatial.0/Conv",
+   "type": "/encoder/stem/spatial/spatial.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stem/fuse/fuse.0/Conv",
+   "type": "/encoder/stem/fuse/fuse.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat0/lat0.0/Conv",
+   "type": "/decoder/lat0/lat0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.0/stages.0.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.0/stages.0.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.0/stages.0.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.0/stages.0.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.0/stages.0.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.0/stages.0.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.0/stages.0.1/Add",
+   "type": "/encoder/stages.0/stages.0.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat.0/lat.0.0/Conv",
+   "type": "/decoder/lat.0/lat.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.1/stages.1.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.1/stages.1.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.1/stages.1.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.1/stages.1.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.1/stages.1.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.1/stages.1.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.1/stages.1.1/Add",
+   "type": "/encoder/stages.1/stages.1.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat.1/lat.1.0/Conv",
+   "type": "/decoder/lat.1/lat.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.2/ReduceMean",
+   "type": "/encoder/stages.2/stages.2.1/net/net.2/ReduceMean",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.2/f/f.0/Conv",
+   "type": "/encoder/stages.2/stages.2.1/net/net.2/f/f.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.2/f/f.2/Conv",
+   "type": "/encoder/stages.2/stages.2.1/net/net.2/f/f.2/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.2/f/f.3/HardSigmoid",
+   "type": "/encoder/stages.2/stages.2.1/net/net.2/f/f.3/HardSigmoid",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.2/Mul",
+   "type": "/encoder/stages.2/stages.2.1/net/net.2/Mul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.1/Add",
+   "type": "/encoder/stages.2/stages.2.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.2/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.2/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.2/ReduceMean",
+   "type": "/encoder/stages.2/stages.2.2/net/net.2/ReduceMean",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.2/f/f.0/Conv",
+   "type": "/encoder/stages.2/stages.2.2/net/net.2/f/f.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.2/f/f.2/Conv",
+   "type": "/encoder/stages.2/stages.2.2/net/net.2/f/f.2/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.2/f/f.3/HardSigmoid",
+   "type": "/encoder/stages.2/stages.2.2/net/net.2/f/f.3/HardSigmoid",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.2/Mul",
+   "type": "/encoder/stages.2/stages.2.2/net/net.2/Mul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.2/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.2/Add",
+   "type": "/encoder/stages.2/stages.2.2/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.2/stages.2.3/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.2/stages.2.3/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.2/ReduceMean",
+   "type": "/encoder/stages.2/stages.2.3/net/net.2/ReduceMean",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.2/f/f.0/Conv",
+   "type": "/encoder/stages.2/stages.2.3/net/net.2/f/f.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.2/f/f.2/Conv",
+   "type": "/encoder/stages.2/stages.2.3/net/net.2/f/f.2/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.2/f/f.3/HardSigmoid",
+   "type": "/encoder/stages.2/stages.2.3/net/net.2/f/f.3/HardSigmoid",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.2/Mul",
+   "type": "/encoder/stages.2/stages.2.3/net/net.2/Mul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.2/stages.2.3/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.2/stages.2.3/Add",
+   "type": "/encoder/stages.2/stages.2.3/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat.2/lat.2.0/Conv",
+   "type": "/decoder/lat.2/lat.2.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.3/stages.3.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.3/stages.3.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.3/stages.3.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.3/stages.3.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.3/stages.3.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.2/ReduceMean",
+   "type": "/encoder/stages.3/stages.3.1/net/net.2/ReduceMean",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.2/f/f.0/Conv",
+   "type": "/encoder/stages.3/stages.3.1/net/net.2/f/f.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.2/f/f.2/Conv",
+   "type": "/encoder/stages.3/stages.3.1/net/net.2/f/f.2/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.2/f/f.3/HardSigmoid",
+   "type": "/encoder/stages.3/stages.3.1/net/net.2/f/f.3/HardSigmoid",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.2/Mul",
+   "type": "/encoder/stages.3/stages.3.1/net/net.2/Mul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.3/stages.3.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.1/Add",
+   "type": "/encoder/stages.3/stages.3.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.0/net.0.0/Conv",
+   "type": "/encoder/stages.3/stages.3.2/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.1/net.1.0/Conv",
+   "type": "/encoder/stages.3/stages.3.2/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.2/ReduceMean",
+   "type": "/encoder/stages.3/stages.3.2/net/net.2/ReduceMean",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.2/f/f.0/Conv",
+   "type": "/encoder/stages.3/stages.3.2/net/net.2/f/f.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.2/f/f.2/Conv",
+   "type": "/encoder/stages.3/stages.3.2/net/net.2/f/f.2/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.2/f/f.3/HardSigmoid",
+   "type": "/encoder/stages.3/stages.3.2/net/net.2/f/f.3/HardSigmoid",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.2/Mul",
+   "type": "/encoder/stages.3/stages.3.2/net/net.2/Mul",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/net/net.3/net.3.0/Conv",
+   "type": "/encoder/stages.3/stages.3.2/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/stages.3/stages.3.2/Add",
+   "type": "/encoder/stages.3/stages.3.2/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.0/net/net.0/net.0.0/Conv",
+   "type": "/encoder/neck.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.0/net/net.1/net.1.0/Conv",
+   "type": "/encoder/neck.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.0/net/net.3/net.3.0/Conv",
+   "type": "/encoder/neck.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.0/Add",
+   "type": "/encoder/neck.0/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.1/net/net.0/net.0.0/Conv",
+   "type": "/encoder/neck.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.1/net/net.1/net.1.0/Conv",
+   "type": "/encoder/neck.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.1/net/net.3/net.3.0/Conv",
+   "type": "/encoder/neck.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.1/Add",
+   "type": "/encoder/neck.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.2/net/net.0/net.0.0/Conv",
+   "type": "/encoder/neck.2/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.2/net/net.1/net.1.0/Conv",
+   "type": "/encoder/neck.2/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.2/net/net.3/net.3.0/Conv",
+   "type": "/encoder/neck.2/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/encoder/neck.2/Add",
+   "type": "/encoder/neck.2/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/lat.3/lat.3.0/Conv",
+   "type": "/decoder/lat.3/lat.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Resize",
+   "type": "/decoder/Resize",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Add",
+   "type": "/decoder/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.2/net/net.0/net.0.0/Conv",
+   "type": "/decoder/ref.2/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.2/net/net.1/net.1.0/Conv",
+   "type": "/decoder/ref.2/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.2/net/net.3/net.3.0/Conv",
+   "type": "/decoder/ref.2/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.2/Add",
+   "type": "/decoder/ref.2/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Resize_1",
+   "type": "/decoder/Resize",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Add_1",
+   "type": "/decoder/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.1/net/net.0/net.0.0/Conv",
+   "type": "/decoder/ref.1/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.1/net/net.1/net.1.0/Conv",
+   "type": "/decoder/ref.1/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.1/net/net.3/net.3.0/Conv",
+   "type": "/decoder/ref.1/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.1/Add",
+   "type": "/decoder/ref.1/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Resize_2",
+   "type": "/decoder/Resize",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Add_2",
+   "type": "/decoder/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.0/net/net.0/net.0.0/Conv",
+   "type": "/decoder/ref.0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.0/net/net.1/net.1.0/Conv",
+   "type": "/decoder/ref.0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.0/net/net.3/net.3.0/Conv",
+   "type": "/decoder/ref.0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref.0/Add",
+   "type": "/decoder/ref.0/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Resize_3",
+   "type": "/decoder/Resize",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/Add_3",
+   "type": "/decoder/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref0/net/net.0/net.0.0/Conv",
+   "type": "/decoder/ref0/net/net.0/net.0.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref0/net/net.1/net.1.0/Conv",
+   "type": "/decoder/ref0/net/net.1/net.1.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref0/net/net.3/net.3.0/Conv",
+   "type": "/decoder/ref0/net/net.3/net.3.0/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref0/Add",
+   "type": "/decoder/ref0/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/decoder/ref0/Add_output_0_DequantizeLinear_Output_converted_QNN_DATATYPE_FLOAT_16",
+   "type": "/decoder/ref0/Add",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "/head/head.1/Conv",
+   "type": "/head/head.1/Conv",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "output_0_0231",
+   "type": "output",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "QNN_DATATYPE_FLOAT_16_converted_output_0_QNN_DATATYPE_FLOAT_32",
+   "type": "QNN",
+   "compute_unit": "NPU"
+  },
+  {
+   "name": "Output",
+   "type": "Output",
+   "compute_unit": "NPU"
+  }
+ ]
+}