deep_sort_pytorch

deep_sort_pytorch copied to clipboard

如果需要训练自己的数据集，我该如何重新训练reID部分。我的YOLO目标检测部分已经训练好了。

2

PiBigStar5712

for track in self.tracks: if not track.is_confirmed(): continue features += track.features targets += [track.track_id for _ in track.features] track.features = [] 在第二帧中，len(features)和len(targets)都为0，到了第三帧就都为84，且里面的数据是有3份相同的。为什么会出现3份相同的数据呢？

goljx

Error while compiling NMS module. Please help!

1

I have tried all possible ways for compiling NMS module. Nothing works! Please help me with this. GCC version- 7.5.0 torch - 1.11.0 torchvision - 0.12.0 python 3.7.16 As per...

soni-H

你好，我把reid模型转成onnx模型后，用onnxruntime推理，处理同一张图片，前处理是一样的； torch的模型输出，前15位： [0.01681411 0.05290637 0.03931383 0.02069167 0.05464995 0.05208057 0.03364787 0.02164895 0.04400286 0.04631684 0.00424759 0.04806569 0.03491891 0.04207694 0.00969397] onnx的模型输出前15位： [1.3163207e-02 3.3286333e-02 3.0172531e-02 2.7824650e-05 5.4625902e-02 2.5197309e-02 4.4952868e-03 5.4582981e-03 2.5899781e-02 3.7061449e-02 1.8202272e-03 1.9214328e-02...

shaqing

https://github.com/ZQPei/deep_sort_pytorch/blob/8cfe2467a4b1f4421ebf9fcbc157921144ffe7cf/deep_sort/deep/model.py#L56

YigenZhong

Bumps [joblib](https://github.com/joblib/joblib) from 0.14.1 to 1.2.0. Changelog Sourced from joblib's changelog. Release 1.2.0 Fix a security issue where eval(pre_dispatch) could potentially run arbitrary code. Now only basic numerics are supported....

dependabot[bot]

Add content and modify bug

7

add some contents about YOLOv5 detector, tracking results showing and modify some errors about kalman_filter.py, detector.YOLOv3.weight.yolo_utils.py(This eliminates the need for compiling NMS operators).

cjliu01

当目标离开镜头再回到镜头，或者发生遮挡object ID会发生变化。 When the target leaves the camera and then returns to the camera, or when occlusion occurs, the object ID will change.

chinanoodles

Fixed question about `deep_sort/deep/resnet.py` mentioned in the [issue(#276)](https://github.com/ZQPei/deep_sort_pytorch/issues/276).

cjliu01

During using it, I found that the performance of using ResNet was not better than the network used in the original paper. Finally, I found that the problem comes from...

cjliu01