search

open-mmlab / mmtracking

OpenMMLab Video Perception Toolbox. It supports Video Object Detection (VID), Multiple Object Tracking (MOT), Single Object Tracking (SOT), Video Instance Segmentation (VIS) with a unified framework.
https://mmtracking.readthedocs.io/en/latest/
Apache License 2.0
3.59k stars 598 forks source link

Problem with inferring custom dataset. #762

Open joey1616 opened 2 years ago

joey1616 commented 2 years ago

Hi, I used mmtracking v1.0.0rc1 to train on my custom dataset. The program I used was strong sort, and the dataset format was MOT. The training process went smoothly and without any errors. However, the program threw errors during the inference process (test and demo_mot_vis). The error thrown during the test process was 1667698819434 The error during the demo_mot_vis process was 1667698865277 I would be grateful for any help or advice.

dyhBUPT commented 2 years ago

Hi, did you change the cfg file?

joey1616 commented 2 years ago

Hi, did you change the cfg file?

Hi, I have changed the cfg file. I used a custom Backbone and Neck. By the way, the custom Backbone and Neck can run without any errors in mmdetection.

dyhBUPT commented 2 years ago

Hi, did you change the cfg file?

Hi, I have changed the cfg file. I used a custom Backbone and Neck. By the way, the custom Backbone and Neck can run without any errors in mmdetection.

Maybe you should use the default cfg first, then see if the error is caused by your changes.

meikorol commented 1 year ago

hello，could you plz tell me how to make my own dataset？whats‘the using software，I find no guidance how to make custom VID dataset