search

detectRecog / PointTrack

PointTrack (ECCV2020 ORAL): Segment as Points for Efficient Online Multi-Object Tracking and Segmentation
Other
261 stars 47 forks source link

Pedestrain training #3

Open iamjeff7 opened 4 years ago

iamjeff7 commented 4 years ago

Dear Author, How can I train for pedestrian instead of car. Thank you!

qa276390 commented 3 years ago

Similar question. @detectRecog will you release the pretrained weights for pedestrian

detectRecog commented 3 years ago

For segmentation, you can train on pedestrians by switching class_id to 2 rather than 1 (for cars) in the dataset. The change of code for tracking should also be similar. There are bugs in the pedestrian model. So we do not plan to provide our trained model for pedestrians. However, for the convenience of comparisons with PointTrack, we have provided the test set results of PointTrack on cars and pedestrians on the GitHub page.

EchoAmor commented 3 years ago

Hi, I want to train on pedestrian too, can u explain more specific? I don't find where to switch class_id, thank you very much! @detectRecog