Add/update model card for seg-large
Browse files
README.md
CHANGED
|
@@ -17,20 +17,19 @@ base_model: roboflow/rfdetr
|
|
| 17 |
|
| 18 |
GGUF-format weights of [Roboflow RF-DETR Seg-Large](https://github.com/roboflow/rf-detr) (segmentation variant) for use with [rfdetr.cpp](https://github.com/mudler/rf-detr.cpp), a C++/ggml implementation that matches the upstream PyTorch model on CPU.
|
| 19 |
|
| 20 |
-
This repo contains all four standard quantizations of this variant. **F16 is the recommended default** β
|
| 21 |
|
| 22 |
## Available files
|
| 23 |
|
| 24 |
-
| File | Quant | Size (MB) |
|
| 25 |
-
|---|---|---:|
|
| 26 |
-
| `rfdetr-seg-large-f32.gguf` | F32 |
|
| 27 |
-
| `rfdetr-seg-large-f16.gguf` β **recommended** | F16 |
|
| 28 |
-
| `rfdetr-seg-large-q8_0.gguf` | Q8_0 |
|
| 29 |
-
| `rfdetr-seg-large-q4_K.gguf` | Q4_K |
|
| 30 |
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
All accuracy numbers are computed against the upstream PyTorch reference (`rfdetr 1.7.0`) on 7 COCO val2017 images at threshold 0.5. Latency is measured with `rfdetr-cli bench` (8 iters + 3 warmup) at T=8 threads on a single AMD Ryzen 9 9950X3D image (`coco_kitchen.jpg`, 640x427).
|
| 34 |
|
| 35 |
## Architecture
|
| 36 |
|
|
@@ -44,17 +43,17 @@ All accuracy numbers are computed against the upstream PyTorch reference (`rfdet
|
|
| 44 |
|
| 45 |
## Quantization notes
|
| 46 |
|
| 47 |
-
- **F32** β full-precision
|
| 48 |
-
- **F16** β matmul-multiplicand weights only; LayerNorms, conv kernels, embeddings, biases, and layer-scale gammas stay F32.
|
| 49 |
-
- **Q8_0** β best size/accuracy tradeoff under F16; ~3Γ smaller than F32 with effectively identical detections.
|
| 50 |
-
- **Q4_K** β smallest practical quant. Rows with `ne[0] % 256 != 0` (the decoder's 128-dim MLP halves
|
| 51 |
|
| 52 |
## Usage
|
| 53 |
|
| 54 |
```bash
|
| 55 |
# 1. Clone + build rfdetr.cpp
|
| 56 |
git clone https://github.com/mudler/rf-detr.cpp
|
| 57 |
-
cd
|
| 58 |
cmake -B build -DRFDETR_BUILD_CLI=ON && cmake --build build -j
|
| 59 |
|
| 60 |
# 2. Download a quant (F16 recommended)
|
|
@@ -71,7 +70,7 @@ build/bin/rfdetr-cli detect \
|
|
| 71 |
|
| 72 |
## Accuracy methodology
|
| 73 |
|
| 74 |
-
All accuracy metrics are computed against the upstream PyTorch reference (rfdetr 1.
|
| 75 |
|
| 76 |
Mask metrics are pixel-wise IoU between binary masks at the **original** image resolution (not the network's working resolution), after sigmoid + bicubic upsample of the per-query mask logits. **Pixel agreement** is the fraction of pixels where the C++ and PyTorch binary masks match.
|
| 77 |
|
|
|
|
| 17 |
|
| 18 |
GGUF-format weights of [Roboflow RF-DETR Seg-Large](https://github.com/roboflow/rf-detr) (segmentation variant) for use with [rfdetr.cpp](https://github.com/mudler/rf-detr.cpp), a C++/ggml implementation that matches the upstream PyTorch model on CPU.
|
| 19 |
|
| 20 |
+
This repo contains all four standard quantizations of this variant. **F16 is the recommended default** β near-identical accuracy to F32 (see the table below), 1.86Γ smaller than F32, and it takes ggml's F32ΓF16 matmul fast path.
|
| 21 |
|
| 22 |
## Available files
|
| 23 |
|
| 24 |
+
| File | Quant | Size (MB) | Recall @ IoU 0.5 | Recall @ IoU 0.95 | Mean mask IoU | Pixel agreement | Latency (median ms, T=8) |
|
| 25 |
+
|---|---|---:|---:|---:|---:|---:|---:|
|
| 26 |
+
| `rfdetr-seg-large-f32.gguf` | F32 | 134.6 | 0.9762 | 0.9762 | 0.9982 | 0.9999 | β |
|
| 27 |
+
| `rfdetr-seg-large-f16.gguf` β **recommended** | F16 | 72.3 | 1.0000 | 0.9762 | 0.9977 | 0.9999 | β |
|
| 28 |
+
| `rfdetr-seg-large-q8_0.gguf` | Q8_0 | 43.1 | 0.9881 | 0.9881 | 0.9943 | 0.9999 | β |
|
| 29 |
+
| `rfdetr-seg-large-q4_K.gguf` | Q4_K | 34.3 | 0.9238 | 0.6119 | 0.9729 | 0.9994 | β |
|
| 30 |
|
| 31 |
+
All accuracy numbers are computed against the upstream PyTorch reference (`rfdetr 1.9.0`) on 7 COCO val2017 images at threshold 0.5.
|
| 32 |
+
No latency benchmark has been recorded for this variant yet, so the latency column is left empty. Run `scripts/quick_bench.sh` locally for timings on your own hardware.
|
|
|
|
| 33 |
|
| 34 |
## Architecture
|
| 35 |
|
|
|
|
| 43 |
|
| 44 |
## Quantization notes
|
| 45 |
|
| 46 |
+
- **F32** β the full-precision conversion, 135 MB, and the closest match to the PyTorch reference (see the table above for measured agreement).
|
| 47 |
+
- **F16** β matmul-multiplicand weights only; LayerNorms, conv kernels, embeddings, biases, and layer-scale gammas stay F32. Accuracy tracks F32 closely on this model, and it takes ggml's F32ΓF16 matmul fast path.
|
| 48 |
+
- **Q8_0** β best size/accuracy tradeoff under F16; ~3.1Γ smaller than F32 with effectively identical detections.
|
| 49 |
+
- **Q4_K** β smallest practical quant. Rows with `ne[0] % 256 != 0` (the decoder's 128-dim MLP halves) silently fall back to Q8_0 per ggml's quantizer logic β net compression is still ~3.9Γ over F32. Use only when the size budget is tight; expect a measurable Recall@0.95 drop relative to F16/Q8_0 (see file table above).
|
| 50 |
|
| 51 |
## Usage
|
| 52 |
|
| 53 |
```bash
|
| 54 |
# 1. Clone + build rfdetr.cpp
|
| 55 |
git clone https://github.com/mudler/rf-detr.cpp
|
| 56 |
+
cd rf-detr.cpp
|
| 57 |
cmake -B build -DRFDETR_BUILD_CLI=ON && cmake --build build -j
|
| 58 |
|
| 59 |
# 2. Download a quant (F16 recommended)
|
|
|
|
| 70 |
|
| 71 |
## Accuracy methodology
|
| 72 |
|
| 73 |
+
All accuracy metrics are computed against the upstream PyTorch reference (`rfdetr 1.9.0`) on 7 COCO val2017 images at threshold 0.5. Each detection match uses greedy Hungarian-style assignment by IoU (β₯ 0.5 lenient, β₯ 0.95 strict) with class equality required.
|
| 74 |
|
| 75 |
Mask metrics are pixel-wise IoU between binary masks at the **original** image resolution (not the network's working resolution), after sigmoid + bicubic upsample of the per-query mask logits. **Pixel agreement** is the fraction of pixels where the C++ and PyTorch binary masks match.
|
| 76 |
|