DETR (ResNet-50 / ResNet-101) (fastforks)

DETR (Carion et al. 2020, Facebook): end-to-end detection with a ResNet backbone and a 6+6 layer transformer encoder-decoder over 100 object queries, no NMS; boxes, scores and COCO classes at the input size. The portable module() is the whole network (class logits and normalised boxes).

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/detr at 29901c51d7fe.

Task: detection. Tier: stable. Outputs: boxes, scores, labels. 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.

variant	weights (author link)	SHA-256	weights licence	commercial use
detr-r101-coco@1	detr-r101-2c7b67e5.pth	2c7b67e52d2e	Apache-2.0	yes
detr-r50-coco@1	detr-r50-e632da11.pth	e632da11ec76	Apache-2.0	yes

Licence

Original-code strings, recorded as they are (research-use package; check them before any other use). Code: Apache-2.0. Files: LICENSE.

Training data: COCO 2017 (train, instance annotations), ImageNet-1k (ILSVRC 2012 classification) (COCO 2017 train (500 epochs); the ResNet backbone was initialised from torchvision's ImageNet-1k model).

Licence evidence: https://github.com/facebookresearch/detr/blob/29901c51d7fe8712168b8d0d64351170bc0f83e0/README.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): detr-r101-coco 3000x4000 (0.6 GiB peak allocated, 0.7 reserved); detr-r50-coco 3000x4000 (0.6 GiB peak allocated, 0.6 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
detr-r101-coco	78.9	80.8	18.69	13.69	5.8x	1.4x	531 / 650	316 / 368	254 / 328
detr-r50-coco	55.5	62.0	12.36	9.93	5.6x	1.2x	459 / 558	280 / 332	218 / 272

Startup and first call

model	load (s): original	portable	kernels	first call (s): original	portable	kernels
detr-r101-coco	0.95	0.81	0.22†	0.06	0.02	0.01
detr-r50-coco	2.55	0.70	0.28†	0.83	0.44	0.03

Fixed-size speed and peak memory, all models

model	mode	640x480	1280x720	1920x1080	12 MP 4000x3000
detr-r101-coco	portable	14.4 ms / 278 / 332	17.6 ms / 307 / 356	18.3 ms / 320 / 416	23.8 ms / 636 / 676
detr-r101-coco	kernels	10.5 ms / 228 / 278	13.0 ms / 248 / 294	13.3 ms / 262 / 354	19.8 ms / 637 / 678
detr-r50-coco	portable	9.5 ms / 241 / 296	11.6 ms / 270 / 320	12.2 ms / 283 / 380	18.7 ms / 599 / 640
detr-r50-coco	kernels	7.7 ms / 191 / 240	8.9 ms / 211 / 256	9.8 ms / 225 / 316	16.7 ms / 600 / 640

Fixed-size cells: time (ms) / peak allocated / peak reserved memory (MiB).

Hardware, protocol and caveats: benchmarks (RTX 5060 Ti 16 GB, sm_120). Original = the authors' code 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

Parity with the original code

Every variant is checked against golden 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| <= 0.2 px and p99 <= 1.0 px.

Usage

from fastforks import detection

model = detection.load("detr-r101-coco")  # portable PyTorch on the GPU if there is one, else the CPU
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: detr-r101-coco, detr-r50-coco.

Kernels backend

detection.load("detr-r101-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.