Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,51 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- en
|
| 4 |
+
- zh
|
| 5 |
+
tags:
|
| 6 |
+
- coreml
|
| 7 |
+
- depth-estimation
|
| 8 |
+
- panoramic
|
| 9 |
+
- ane
|
| 10 |
+
- apple-neural-engine
|
| 11 |
+
- quantized
|
| 12 |
+
- int8
|
| 13 |
+
- ios
|
| 14 |
+
license: apache-2.0
|
| 15 |
+
pipeline_tag: depth-estimation
|
| 16 |
+
---
|
| 17 |
+
|
| 18 |
+
# DAP (Depth Anything Panoramic) - 256x512 INT8 CoreML Model
|
| 19 |
+
|
| 20 |
+
This repository contains the quantized CoreML model for **Depth Anything Panoramic (DAP)**, tailored and optimized for Apple Neural Engine (ANE) on iOS devices (e.g. iPhone 15 / iPhone 16 series).
|
| 21 |
+
|
| 22 |
+
## Model Overview
|
| 23 |
+
|
| 24 |
+
- **Architecture**: Depth Anything V2 Panoramic Backbone (DAP)
|
| 25 |
+
- **Input Resolution**: `256 x 512` (Equirectangular Panoramic Frame, RGB, 3 channels)
|
| 26 |
+
- **Quantization**: INT8 Linear Weight Quantization
|
| 27 |
+
- **Runtime Acceleration**: Apple Neural Engine (ANE) + MPSGraph
|
| 28 |
+
- **Package Format**: Apple CoreML `.mlpackage`
|
| 29 |
+
- **Compressed Archive**: `dap_256x512_int8.mlpackage.zip` (277 MB)
|
| 30 |
+
|
| 31 |
+
## Benchmark Performance on Hardware (iPhone 15)
|
| 32 |
+
|
| 33 |
+
| Metric | Measured Value |
|
| 34 |
+
| :--- | :--- |
|
| 35 |
+
| **Engine / Compute Unit** | Apple Neural Engine (`.all`) |
|
| 36 |
+
| **Average Inference Latency** | **82.5 ms** (~12.1 FPS) |
|
| 37 |
+
| **Cold Start / Load Latency** | ~650 ms |
|
| 38 |
+
| **Runtime Memory Peak** | **410 MB** |
|
| 39 |
+
| **Output Metric** | Relative / Metric Depth Map (`256 x 512`, FP32) |
|
| 40 |
+
|
| 41 |
+
## Quickstart (Download & Extract)
|
| 42 |
+
|
| 43 |
+
You can download and extract the model into your WalkMate project or custom iOS application using:
|
| 44 |
+
|
| 45 |
+
```bash
|
| 46 |
+
# Download
|
| 47 |
+
curl -L -o dap_256x512_int8.mlpackage.zip https://huggingface.co/Sysking/dap-quantized/resolve/main/dap_256x512_int8.mlpackage.zip
|
| 48 |
+
|
| 49 |
+
# Extract
|
| 50 |
+
unzip dap_256x512_int8.mlpackage.zip
|
| 51 |
+
```
|