| --- |
| license: apache-2.0 |
| library_name: rfdetr.cpp |
| tags: |
| - object-detection |
| - rfdetr |
| - gguf |
| - ggml |
| - cpp-inference |
| pipeline_tag: object-detection |
| base_model: roboflow/rfdetr |
| --- |
| |
| # RF-DETR Large — GGUF for rfdetr.cpp |
|
|
| GGUF-format weights of [Roboflow RF-DETR Large](https://github.com/roboflow/rf-detr) (detection 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. |
|
|
| 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 \|Δscore\| | Latency (median ms, T=8) | |
| |---|---|---:|---:|---:|---:|---:| |
| | `rfdetr-large-f32.gguf` | F32 | 125.9 | 0.9731 | 0.9621 | 0.0071 | 236.6 | |
| | `rfdetr-large-f16.gguf` ← **recommended** | F16 | 68.2 | 0.9731 | 0.9731 | 0.0070 | 237.1 | |
| | `rfdetr-large-q8_0.gguf` | Q8_0 | 41.1 | 0.9731 | 0.9463 | 0.0092 | 251.2 | |
| | `rfdetr-large-q4_K.gguf` | Q4_K | 33.4 | 0.9573 | 0.8152 | 0.0208 | 272.0 | |
| |
| 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: 704×704 |
| - Patch size: 14 |
| - Decoder layers: 4 |
| - Object queries: 300 |
| - Task: object detection (boxes only) |
|
|
| ## 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 |
|
|
| ```bash |
| # 1. Clone + build rfdetr.cpp |
| git clone https://github.com/mudler/rf-detr.cpp |
| cd rt-detr.cpp |
| cmake -B build -DRFDETR_BUILD_CLI=ON && cmake --build build -j |
| |
| # 2. Download a quant (F16 recommended) |
| hf download mudler/rfdetr-cpp-large rfdetr-large-f16.gguf --local-dir models/ |
| |
| # 3. Run detection |
| build/bin/rfdetr-cli detect \ |
| --model models/rfdetr-large-f16.gguf \ |
| --input my_image.jpg \ |
| --threshold 0.5 --threads 8 \ |
| --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. |
|
|
| See [BENCHMARK.md](https://github.com/mudler/rf-detr.cpp/blob/main/BENCHMARK.md) and [`benchmarks/results/accuracy_sweep.json`](https://github.com/mudler/rf-detr.cpp/blob/main/benchmarks/results/accuracy_sweep.json) for the full sweep across all 32 (variant × quant) cells. |
|
|
| ## License |
|
|
| Apache-2.0 — matches the upstream [rfdetr](https://github.com/roboflow/rf-detr) license. |
|
|
