RF-DETR Seg-Small — GGUF for rfdetr.cpp
GGUF-format weights of
Roboflow RF-DETR Seg-Small (segmentation variant) for use with
rfdetr.cpp, a C++/ggml implementation that matches the upstream PyTorch model on CPU.
This repo contains all four standard quantizations of this variant. F16 is the recommended default — same accuracy as F32, 1.85× smaller, and typically the fastest on modern CPUs thanks to ggml's F32×F16 matmul fast path.
Available files
| File | Quant | Size (MB) | Recall @ IoU 0.5 | Recall @ IoU 0.95 | Mean mask IoU | Pixel agreement | Latency (median ms, T=8) |
|---|
rfdetr-seg-small-f32.gguf | F32 | 127.6 | 1.0000 | 1.0000 | 0.9924 | 0.9998 | 155.2 |
rfdetr-seg-small-f16.gguf ← recommended | F16 | 68.3 | 1.0000 | 1.0000 | 0.9925 | 0.9999 | 150.2 |
rfdetr-seg-small-q8_0.gguf | Q8_0 | 40.4 | 0.9821 | 0.9821 | 0.9912 | 0.9998 | 148.8 |
rfdetr-seg-small-q4_K.gguf | Q4_K | 32.4 | 0.9533 | 0.3761 | 0.9665 | 0.9989 | 171.2 |
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).
Architecture
- Backbone: DINOv2-small
- Input resolution: 384×384
- Patch size: 12
- Decoder layers: 4
- Object queries: 100
- Task: instance segmentation (boxes + per-query masks)
- Mask resolution: 96×96 per query (image_size / 4)
Quantization notes
- F32 — full-precision reference, ~120 MB. Bit-exact PyTorch parity.
- F16 — matmul-multiplicand weights only; LayerNorms, conv kernels, embeddings, biases, and layer-scale gammas stay F32. Lossless on this model and consistently the fastest variant on CPU.
- Q8_0 — best size/accuracy tradeoff under F16; ~3× smaller than F32 with effectively identical detections.
- Q4_K — smallest practical quant. Rows with
ne[0] % 256 != 0 (the decoder's 128-dim MLP halves, 60 tensors) silently fall back to Q8_0 per ggml's quantizer logic — net compression is still ~3.8× 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).
Usage
1# 1. Clone + build rfdetr.cpp
2git clone https://github.com/mudler/rf-detr.cpp
3cd rt-detr.cpp
4cmake -B build -DRFDETR_BUILD_CLI=ON && cmake --build build -j
5
6# 2. Download a quant (F16 recommended)
7hf download mudler/rfdetr-cpp-seg-small rfdetr-seg-small-f16.gguf --local-dir models/
8
9# 3. Run segmentation (writes per-detection PNG masks to /tmp/seg_masks/)
10build/bin/rfdetr-cli detect \
11 --model models/rfdetr-seg-small-f16.gguf \
12 --input my_image.jpg \
13 --threshold 0.5 --threads 8 \
14 --masks /tmp/seg_masks \
15 --output detections.json
Accuracy methodology
All accuracy metrics are computed against the upstream PyTorch reference (rfdetr 1.7.0) on 7 COCO val2017 images at threshold 0.5. Each detection match uses greedy Hungarian-style assignment by IoU (≥ 0.5 lenient, ≥ 0.95 strict) with class equality required.
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.
See
BENCHMARK.md and
benchmarks/results/accuracy_sweep.json for the full sweep across all 32 (variant × quant) cells.
License
Apache-2.0 — matches the upstream
rfdetr license.
