Badminton analysis — browser models

ONNX models used by the in-browser badminton analysis app (onnxruntime-web, WebGPU / WASM). The video never leaves the viewer's device; this repo only hosts the weights.

File	What	Source / license
`tracknet3_fp32.onnx`, `tracknet3_fp16.onnx`	Shuttlecock heatmaps. Input `(1, 32, H, W)`: median background + 8 frames, RGB / 255, padded from 27 to 32 channels with zeros (onnxruntime-web 1.30 WebGPU computes a 27-channel first conv wrongly). BatchNorm folded into the convs. Output `(1, 8, H, W)`.	TrackNetV3, Chen & Wang, ACM MMAsia 2023 — qaz812345/TrackNetV3, MIT (code and weights)
`rtmo_s.onnx`	One-stage multi-person pose (17 COCO keypoints), input `(1, 3, 640, 640)` BGR 0–255, letterboxed with 114.	RTMO, OpenMMLab mmpose, Apache-2.0
`rtmo_m_fp16.onnx`, `rtmo_l_fp16.onnx`	Larger RTMO variants (FP16, inputs/outputs FP32).	RTMO, OpenMMLab mmpose, Apache-2.0

Exported with PyTorch 2.11 / opset 17; FP16 versions via onnxconverter_common.float16 with FP32 I/O.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support