YOLOX (tiny / S / M / L / X) (fastforks)
YOLOX (Ge et al. 2021, Megvii): anchor-free one-stage detector, all convolutions: a CSPDarknet backbone, a PAFPN neck and a decoupled head that predicts, per cell of the stride 8 / 16 / 32 maps, a box (centre and size), an objectness and a score per class; a cell's score is objectness x its best class; class-wise NMS. COCO checkpoints (80 classes) from the authors. The image is fitted into 640 x 640 (tiny: 416 x 416); vendored from the reference (no yolox dependency). The portable module() is the whole network (decoded boxes, scores and labels per cell, before NMS).
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/Megvii-BaseDetection/YOLOX at 6ddff4824372.
Task: detection. Tier: experimental. 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 |
|---|---|---|---|---|
yolox-l@1 |
yolox_l.pth | 1e6b7fa62403 |
Apache-2.0 | yes |
yolox-m@1 |
yolox_m.pth | 60076992b32d |
Apache-2.0 | yes |
yolox-s@1 |
yolox_s.pth | f55ded7181e1 |
Apache-2.0 | yes |
yolox-tiny@1 |
yolox_tiny.pth | 9de513de589a |
Apache-2.0 | yes |
yolox-x@1 |
yolox_x.pth | 5652330b6ae8 |
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. Licence files: LICENSE.
Training data: COCO 2017 (train, instance annotations) (COCO 2017 train, 300 epochs from scratch (no ImageNet pretraining; Ge et al. 2021, section 2.1)).
Licence evidence: https://github.com/Megvii-BaseDetection/YOLOX/blob/6ddff4824372906469a7fae2dc3206c7aa4bbaee/LICENSE (Apache-2.0); https://github.com/Megvii-BaseDetection/YOLOX/releases/tag/0.1.1rc0 (the weights)
Input size
any H x W >= 32, padded to a multiple of 32
Tested up to (portable, fp16, RTX 5060 Ti 16 GB): yolox-l 3000x4000 (0.6 GiB peak allocated, 0.6 reserved); yolox-m 3000x4000 (0.5 GiB peak allocated, 0.6 reserved); yolox-s 3000x4000 (0.5 GiB peak allocated, 0.5 reserved); yolox-tiny 3000x4000 (0.5 GiB peak allocated, 0.5 reserved); yolox-x 3000x4000 (0.7 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 |
|---|---|---|---|---|---|---|---|---|---|
yolox-l |
β | 15.1 | 8.13 | 5.02 | 3.0x | 1.6x | 265 / 298 | 165 / 206 | 139 / 250 |
yolox-m |
β | 12.4 | 6.42 | 3.45 | 3.6x | 1.9x | 142 / 160 | 110 / 132 | 77 / 184 |
yolox-s |
β | 10.4 | 5.55 | 2.43 | 4.3x | 2.3x | 67 / 90 | 79 / 100 | 40 / 92 |
yolox-tiny |
β | 9.79 | 5.58 | 1.80 | 5.4x | 3.1x | 32 / 58 | 60 / 92 | 22 / 88 |
yolox-x |
β | 17.2 | 11.8 | 7.82 | 2.2x | 1.5x | 450 / 504 | 255 / 300 | 258 / 358 |
Startup and first call
| model | load (s): original | portable | kernels | first call (s): original | portable | kernels |
|---|---|---|---|---|---|---|
yolox-l |
0.65 | 0.41 | 1.72 | 0.03 | 1.16 | 0.36 |
yolox-m |
0.32 | 0.24 | 1.60 | 0.03 | 0.36 | 0.19 |
yolox-s |
0.22 | 0.15 | 1.55 | 0.03 | 1.35 | 0.30 |
yolox-tiny |
1.55 | 0.18 | 1.58 | 0.52 | 1.31 | 0.31 |
yolox-x |
0.76 | 0.59 | 2.06 | 0.08 | 0.21 | 0.26 |
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) |
|---|---|---|---|---|---|---|---|---|---|---|---|
yolox-l |
800x1344 | 8.13 ms | 6.15 ms | 3.10 ms | 5.02 ms | 15.3 s | 21.2 s | 165 / 206 | 173 / 208 | 139 / 250 | 279 / 324 |
yolox-m |
800x1344 | 6.42 ms | 3.70 ms | 2.03 ms | 3.45 ms | 13.4 s | 20.1 s | 110 / 132 | 111 / 132 | 77 / 184 | 152 / 194 |
yolox-s |
800x1344 | 5.55 ms | 2.95 ms | 1.02 ms | 2.43 ms | 12.7 s | 17.2 s | 79 / 100 | 74 / 96 | 40 / 92 | 86 / 126 |
yolox-tiny |
800x1344 | 5.58 ms | 2.87 ms | 0.561 ms | 1.80 ms | 13.1 s | 23.7 s | 60 / 92 | 58 / 94 | 22 / 88 | 62 / 90 |
yolox-x |
800x1344 | 11.8 ms | 11.3 ms | 5.28 ms | 7.82 ms | 15.8 s | 24.1 s | 255 / 300 | 265 / 296 | 258 / 358 | 432 / 478 |
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 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 |
|---|---|---|---|---|---|---|
yolox-tiny |
COCO val2017, every 25th image (200), box mAP | 38.16 | 38.20 | 38.21 | 38.29 | |
yolox-s |
COCO val2017, every 25th image (200), box mAP | 44.67 | 44.75 | 44.67 | 44.73 | |
yolox-m |
COCO val2017, every 25th image (200), box mAP | 50.34 | 50.28 | 50.29 | 50.38 | |
yolox-l |
COCO val2017, every 25th image (200), box mAP | 54.13 | 54.11 | 54.13 | 54.19 | |
yolox-x |
COCO val2017, every 25th image (200), box mAP | 54.64 | 54.61 | 54.64 | 54.60 |
Measured on NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks db9e79d5, 2026-11-13.
Parity with the original code
Every variant is checked against reference outputs of the original code at its pinned commit. CPU, float32: max |diff| <= 0.00037 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.2 and p99 <= 0.51 (absolute error of the output boxes).
Limitations
- The image is resized like OpenCV 4's 8-bit INTER_LINEAR (what the reference calls); OpenCV 5 resizes differently by up to two grey levels, so a reference run on OpenCV 5 differs slightly.
- Boxes are not clamped to the image, as upstream: at the right or bottom border they can reach a few pixels into the padding.
- One label per box (the cell's best class), as upstream: a cell does not report a second class.
Usage
from fastforks import detection
model = detection.load("yolox-l") # 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: yolox-l, yolox-m, yolox-s, yolox-tiny, yolox-x.
Kernels backend
detection.load("yolox-l", mode="kernels") runs our CUDA kernels (cudnn_hacc, fvm_conv, fvm_extra, fvm_mb, fvm_r3). 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.