rcnn / README.md
Flor's picture
card: update from fastforks 1cbf1671
364cca5 verified
|
Raw History Blame Contribute Delete
7.37 kB
---
library_name: "fastforks"
pipeline_tag: "object-detection"
license: "cc-by-sa-3.0"
tags:
- "detection"
- "object-detection"
- "instance-segmentation"
- "cuda"
- "pytorch"
- "fastforks"
- "fastforks-card"
---
# detectron2 Mask / Faster R-CNN (R50-FPN) (fastforks)
Mask R-CNN (He et al. 2017) and Faster R-CNN (Ren et al. 2015) from the detectron2 model zoo, COCO, ResNet-50 + FPN backbone, RPN proposals, ROIAlign box head (+ mask head); returns boxes, classes and instance masks at the input size. The portable `module()` is backbone + FPN.
This repo is the model card only: it holds no weights. fastforks downloads the authors' weights from the links below and checks them against the SHA-256 in the model card. Code: <https://github.com/kleinicke/fastforks>. Kernels for the `kernels` backend: <https://huggingface.co/fastforks/kernels>. Original code: <https://github.com/facebookresearch/detectron2> at `1e3e13b`.
Task: detection. Tier: stable. Outputs: `boxes`, `scores`, `labels`, `masks`. Backends: `portable`, `kernels`.
Input: one RGB image. Output: boxes, class labels, scores and instance masks.
## Variants
Weights are the authors' files (not hosted here). SHA-256 shortened to 12 hex digits; the full hashes are in [`card.toml`](https://github.com/kleinicke/fastforks/blob/main/src/fastforks/detection/models/rcnn/card.toml).
| variant | weights (author link) | SHA-256 | weights licence | commercial use |
|---|---|---|---|---|
| `faster-rcnn-r50-fpn-3x-coco@1` | [faster_rcnn_R_50_FPN_3x.pkl](https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_R_50_FPN_3x/137849458/model_final_280758.pkl) | `3c25caca37ba` | CC-BY-SA-3.0 | yes |
| `mask-rcnn-r50-fpn-3x-coco@1` | [mask_rcnn_R_50_FPN_3x.pkl](https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x/137849600/model_final_f10217.pkl) | `9a737e290372` | CC-BY-SA-3.0 | yes |
## Licence
Original-code strings, recorded as they are (research-use package; check them before any other use). Code: Apache-2.0. Licence files: `LICENSE`.
Training data: COCO 2017 (train, instance annotations) (COCO 2017 train (3x schedule); the ResNet-50 backbone was initialised from MSRA's ImageNet-1k model).
- Licence note: The detectron2 model-zoo weights are CC-BY-SA-3.0 (share-alike), and the COCO images carry their own licences.
Licence evidence: https://github.com/facebookresearch/detectron2/blob/1e3e13bbf607b54f62205c4c33922521822fb298/MODEL_ZOO.md (License section)
## Input size
any H x W >= 32, padded to a multiple of 32
Tested up to (portable, fp16, RTX 5060 Ti 16 GB): `faster-rcnn-r50-fpn-3x-coco` 3000x4000 (0.6 GiB peak allocated, 0.7 reserved); `mask-rcnn-r50-fpn-3x-coco` 3000x4000 (0.6 GiB peak allocated, 0.7 reserved)
## Speed and memory
### Detection, synthetic, 800x1344
| model | time (ms): original fp32 | original | portable | kernels | kernels speed-up: vs original | vs portable | peak memory (MiB, allocated / reserved): original | portable | kernels |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| `faster-rcnn-r50-fpn-3x-coco` | 83.6 | 41.4 | 22.8 | 19.5 | 2.1x | 1.2x | 626 / 708 | 390 / 490 | 391 / 450 |
| `mask-rcnn-r50-fpn-3x-coco` | 85.4 | 43.5 | 24.6 | 21.0 | 2.1x | 1.2x | 604 / 650 | 352 / 440 | 353 / 400 |
### Startup and first call
| model | load (s): original | portable | kernels | first call (s): original | portable | kernels |
| --- | --- | --- | --- | --- | --- | --- |
| `faster-rcnn-r50-fpn-3x-coco` | 0.73 | 0.53 | 0.14† | 0.04 | 0.25 | 0.04 |
| `mask-rcnn-r50-fpn-3x-coco` | 0.97 | 1.50 | 0.22† | 0.84 | 0.42 | 0.04 |
### torch.compile and TensorRT
| model | input | time: portable | compile | TensorRT fp16 | kernels | preparation: compile (cold first call) | ONNX + engine build | peak MiB (alloc / reserved): portable | compile | kernels | TensorRT (engine + torch) |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| `faster-rcnn-r50-fpn-3x-coco` | 800x1344 | 22.8 ms | 18.9 ms | 5.35 ms (backbone + FPN only) | 19.5 ms | 26.0 s | 15.6 s | 390 / 490 | 413 / 438 | 391 / 450 | 439 / 494 |
| `mask-rcnn-r50-fpn-3x-coco` | 800x1344 | 24.6 ms | 21.5 ms | 5.34 ms (backbone + FPN only) | 21.0 ms | 29.3 s | 18.3 s | 352 / 440 | 333 / 358 | 353 / 400 | 444 / 496 |
### Fixed-size speed and peak memory, all models
| model | mode | 480x640 | 720x1280 | 1080x1920 | 12 MP 3000x4000 |
| --- | --- | --- | --- | --- | --- |
| `faster-rcnn-r50-fpn-3x-coco` | portable | 18.5 ms / 309 / 420 | 21.5 ms / 338 / 426 | 22.2 ms / 352 / 454 | 27.2 ms / 602 / 734 |
| `faster-rcnn-r50-fpn-3x-coco` | kernels | 16.1 ms / 310 / 422 | 18.7 ms / 339 / 460 | 18.4 ms / 353 / 488 | 24.6 ms / 602 / 736 |
| `mask-rcnn-r50-fpn-3x-coco` | portable | 18.1 ms / 314 / 422 | 23.1 ms / 348 / 440 | 25.1 ms / 413 / 536 | 27.1 ms / 607 / 736 |
| `mask-rcnn-r50-fpn-3x-coco` | kernels | 15.6 ms / 315 / 422 | 19.6 ms / 349 / 472 | 22.3 ms / 414 / 568 | 24.6 ms / 607 / 736 |
Fixed-size cells: time (ms) / peak allocated / peak reserved memory (MiB).
Hardware, protocol and caveats: [benchmarks](https://github.com/kleinicke/fastforks/blob/main/docs/benchmarks.md) (RTX 5060 Ti 16 GB, sm_120). Original = the original code named above at the precision of portable (fp16, or `(fp32)` where fp16 is unsupported or not finite); original fp32 = its default fp32 run. Peak memory is the torch allocator's.
## Accuracy
### On public datasets
pycocotools COCOeval of the top 100 detections per image. Original = the original code named under Details, same checkpoint and protocol.
| variant | dataset | original fp32 | original fp16 | portable | kernels | note |
|---|---|---|---|---|---|---|
| `mask-rcnn-r50-fpn-3x-coco` | COCO val2017, every 10th image, box mAP / mask mAP | 42.13 / 38.67 | 42.13 / 38.66 | 42.12 / 38.65 | 42.10 / 38.64 | |
| `faster-rcnn-r50-fpn-3x-coco` | COCO val2017, every 10th image, box mAP | 41.54 | 41.44 | 41.58 | 41.58 | |
Measured on NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 939adfe8, 2026-11-06.
### Parity with the original code
Every variant is checked against reference outputs of the original code at its pinned commit. CPU, float32: max |diff| <= 0.002 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 1.0 and p99 <= 2.0 (absolute error of the output `boxes`).
## Usage
```python
from fastforks import detection
model = detection.load("faster-rcnn-r50-fpn-3x-coco") # mode="auto": the preferred backend from recorded measurements
result = model("photo.jpg")
result.boxes, result.scores, result.labels # [N, 4], [N], [N]
result.backend # what ran, e.g. Backend(mode='portable', device='cuda:0', dtype='float16')
```
Install with `pip install fastforks` (see <https://github.com/kleinicke/fastforks>). Other variants: `faster-rcnn-r50-fpn-3x-coco`, `mask-rcnn-r50-fpn-3x-coco`.
## Kernels backend
`detection.load("faster-rcnn-r50-fpn-3x-coco", mode="kernels")` runs our CUDA kernels (`cudnn_hacc`). It needs a prebuilt kernel for your GPU architecture (downloaded from <https://huggingface.co/fastforks/kernels> on first use) and a recent NVIDIA driver; if there is none it raises `BackendUnavailableError` with the reason and never falls back silently. `mode="auto"` picks `kernels` when it can run and logs the choice.
<!-- staged from fastforks 1cbf1671 (2026-11-15) -->