Liu comments

Results 9 comments of

Liu

你们不能提供个docker镜像吗，这个要求pytorch版本那么低，安装环境各种报错，能不能整一下啊

我觉得一个东西出问题，特别是编译的问题，不用纠结，直接删除跟这个东西有关的所有相关部分（我直接将DCN改成一般的卷积，反正网络的结构和它训练的过程才是最重要的）

How should I operate to get the test score results HOTA MOTA IDF1 MT ML FP FN IDs on the mot17 data set?

You can refer to 'tools/eval_hota.sh'

Will this paper be accepted by any journals this year, or has it not been accepted?

@Bewear777 SparseTrack has not been accepted. The paper is currently under review.

assert img is not None

可以核对一下config文件中的文件路径，例如这里： ![image](https://github.com/hustvl/SparseTrack/assets/93588037/a3a6deb5-400b-48ed-aa34-72aa4853d2b3)

Inference model on video

Thank you for your feedback. We will soon add the script for video inference.

Inference model on video

The code for inference has been updated

How to conduct HOTA MOTA IDF1 MT ML FP FN IDs FPS evaluation

For the results with CLEAR metrics, you can run tracking scripts [such as, mot17_track_cfg.py] to get the results. For the results with HOTA metrics, you can refer to 'tools/eval_hota.sh', modify...

Is there any FPS benchmarks?

With the YOLOX-x detector, SparseTrack achieve 22 FPS on MOT17 test set (800 x 1440) and 12.5 FPS on MOT20 test set (896 x 1600).

CODE ABOUT INTEGRATING DCM ALGORITHM INTO DIFFERENT TRACKERS

We have removed the original Ablation 3) experiment. The main reason is that we found that DCM does not show significant gains for detectors with poor detection performance when the...