UniMatch
UniMatch unifies optical flow, stereo and depth estimation with one transformer matcher (TPAMI 2023); GMFlow is its flow model. LibreYOLO ships the optical flow task with the GMFlow checkpoints, under MIT.
- 작업
- optical flow
- 크기
- s, l at 640 px
- 설치
pip install libreyolo- 지원 티어
- 추론 전용, v1.8.0부터 지원. 예측, 검증, 내보내기만 지원합니다. 학습 기능은 적용되지 않습니다.
- 라이선스
- 코드 MIT, 가중치 MIT. 상업적 사용
GMFlow treats optical flow as a global matching problem: a transformer
compares features of the two frames and a softmax over all positions gives the
match, optionally refined by local regression. The vendored upstream code also
contains the stereo and depth paths, but only flow is wired in LibreYOLO. Pick
l for real-world footage, as upstream recommends.
Install
UniMatch needs no optional extra. Everything it imports is in the base install.
pip install libreyoloPredict
Weights download from the LibreYOLO Hugging Face organization on first use.
The class LibreUniMatch is also exported from libreyolo if you prefer to build the
model directly.
from libreyolo import LibreYOLO model = LibreYOLO("LibreUniMatchl-flow.pt") # weights download on first useresults = model(["frame_0001.jpg", "frame_0002.jpg"]) flow = results[0].flowprint(flow.data.shape) # (H, W, 2) float32, (dx, dy) in pixelsprint(flow.mean_magnitude) # mean displacement in pixelsfrom libreyolo import LibreYOLO model = LibreYOLO("LibreUniMatchl-flow.pt") # N frames give N - 1 results; result i is the flow from frame i to i + 1.for result in model("video.mp4", stream=True, save=True, save_flow=True): print(result.flow.max_magnitude)libreyolo predict model=LibreUniMatchl-flow.pt source=frame_0001.jpg,frame_0002.jpg save=truelibreyolo predict model=LibreUniMatchl-flow.pt source=video.mp4 save=true arrows=true save_flow=trueFlow needs at least two frames: a single image raises. A list of images, a
directory sorted by name, a video or a stream is one sequence, and N frames
give N - 1 results. result.flow is a FlowField with an (H, W, 2) float32
array on the original canvas of the first frame; flow.data[y, x] is
(dx, dy) in pixels, +x right and +y down. UniMatch fills no confidence map, so flow.confidence is None. result.boxes is
None. save=True writes the Middlebury colour wheel, arrows=True or
warp=True pick the other views, and save_flow=True also writes each pair's
raw field as a .flo file. The optical flow task
page describes the payload and views in full.
Variants
| Size | Upstream checkpoint | File |
|---|---|---|
s | gmflow-scale1-things: global matching at 1/8 (Chairs, then Things) | LibreUniMatchs-flow.pt |
l | gmflow-scale2-regrefine6-mixdata: GMFlow+ at two scales with 6 refinement steps (canonical) | LibreUniMatchl-flow.pt |
l | Sintel fine-tune | LibreUniMatchl-flow-sintel.pt |
l | KITTI-2015 fine-tune | LibreUniMatchl-flow-kitti.pt |
Only l has Sintel and KITTI fine-tunes. LibreUniMatch defaults to size
l. Preprocessing reproduces upstream inference_flow: portrait frames are
transposed to landscape, and both frames are resized so each side rounds up to
the size's padding factor (16 for s, 32 for l). The field is resized back
and its vectors rescaled. One deliberate fix over upstream: for portrait frames
upstream leaves dx and dy exchanged after transposing back; LibreYOLO swaps
them so the field follows the flow contract.
Training is not offered: train() raises NotImplementedError.
Validate
val() runs the shared flow validator on a flow dataset (frames per sequence
folder, Middlebury .flo, KITTI 16-bit PNG or .npy ground truth; see the
dataset format). It reports the mean
end-point error metrics/epe, metrics/px1 and metrics/px3 (share of valid
pixels under 1 and 3 px of error) and the KITTI outlier rate
metrics/fl_all.
from libreyolo import LibreYOLO model = LibreYOLO("LibreUniMatchl-flow.pt")metrics = model.val(data="my_flow.yaml") print(metrics["metrics/epe"]) # mean end-point error, pixelsprint(metrics["metrics/px1"]) # share of pixels under 1 px errorprint(metrics["metrics/fl_all"]) # KITTI outlier ratelibreyolo val model=LibreUniMatchl-flow.pt data=my_flow.yamlExport
| 작업 | ONNX | TorchScript | ExecuTorch | TensorRT | OpenVINO | Paddle | MNN | RKNN | Axelera | ncnn | TFLite | CoreML | Core AI |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Optical flow | Optical flow to ONNX: 지원하지 않음 | Optical flow to TorchScript: 지원하지 않음 | Optical flow to ExecuTorch: 지원하지 않음 | Optical flow to TensorRT: 지원하지 않음 | Optical flow to OpenVINO: 지원하지 않음 | Optical flow to Paddle: 지원하지 않음 | Optical flow to MNN: 지원하지 않음 | Optical flow to RKNN: 지원하지 않음 | Optical flow to Axelera: 지원하지 않음 | Optical flow to ncnn: 지원하지 않음 | Optical flow to TFLite: 지원하지 않음 | Optical flow to CoreML: 지원하지 않음 | Optical flow to Core AI: 지원하지 않음 |
Export is not available yet for any format. Optical flow export is blocked
until the two-input graph (image1, image2) and the backend vector-rescale
contract of the flow task are implemented for a family.
Checkpoints
Every published weight file for this family.
| 파일 | 가중치 라이선스 |
|---|---|
| Optical flow | |
| LibreUniMatchs-flow.pt | mit |
| LibreUniMatchl-flow.pt | mit |
| LibreUniMatchl-flow-sintel.pt | mit |
| LibreUniMatchl-flow-kitti.pt | mit |
위의 모든 파일은 현재 LibreYOLO 조직에 있으며 처음 사용할 때 내려받습니다.
Licensing
내려받는 특정 가중치의 Hugging Face 저장소에서 라이선스를 확인하십시오. LibreYOLO 조직의 모든 체크포인트에는 라이선스가 있으며 한 계열 안에서도 항상 같지는 않습니다. 해당 저장소가 신뢰할 수 있는 기준입니다. 아래 요약은 이 페이지를 마지막으로 검증했을 때 적용된 내용을 설명합니다.
관련 라이선스에 관한 설명이며 법률 자문이 아닙니다. 상업적으로 중요한 사안이라면 라이선스를 직접 읽고 별도의 법률 자문을 받으십시오.
- 원작
- UniMatch, Autonomous Vision Group (University of Tübingen)
- 업스트림 라이선스
- MIT
- LibreYOLO 코드
- MIT
- 가중치
- MIT, huggingface.co/LibreYOLO에 다시 게시됨
- 해석
- MIT is a permissive license: the code and weights can be used in commercial and closed-source products, provided the copyright and license notice travel with copies. The upstream unimatch/ package at commit 943a9900 is vendored verbatim under libreyolo/models/unimatch/_vendor/ with its MIT LICENSE; LibreYOLO's wrapper code is also MIT. Upstream publishes the GMFlow checkpoints (gmflow-scale1-things as s, gmflow-scale2-regrefine6-mixdata as l, plus its Sintel and KITTI-2015 fine-tunes) without a separate weight license, so LibreYOLO hosts them on its Hugging Face org under the releasing project's MIT license: an implied grant, disclosed as LibreYOLO's interpretation and not confirmed by the publisher. No gating; only the optical flow task is wired, and training is out of scope.
UniMatch is the work of the
Autonomous Vision Group
(University of Tübingen). The upstream unimatch package is
vendored verbatim under its MIT licence. Upstream publishes the GMFlow
checkpoints without a separate weight licence, so LibreYOLO hosts them under
the releasing project's MIT licence: an implied grant, LibreYOLO's
interpretation and not confirmed by the publisher. No gating. MIT allows
commercial and closed-source use as long as the copyright and licence notice
travel with copies.
Citation
@article{xu2023unifying,
title={Unifying Flow, Stereo and Depth Estimation},
author={Xu, Haofei and Zhang, Jing and Cai, Jianfei and Rezatofighi, Hamid and Yu, Fisher and Tao, Dacheng and Geiger, Andreas},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
year={2023}
}github.com/autonomousvision/unimatch#citation에 있는 저자의 인용 블록에서 복사했습니다.