search

detectRecog / PointTrack

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

How to visualize the MOTS results #12

Closed EchoAmor closed 3 years ago

EchoAmor commented 3 years ago

Thanks for your great work! It's really helpful! But after I got the tracking results ,I didn't find the visualization.py , can u publish the code ? Thanks a lot!

EchoAmor commented 3 years ago

@detectRecog Hi, Is the visualization code in the datasets?

Ilyabasharov commented 3 years ago

@EchoAmor hello, the code for vis is located in datasets (link). so u should run this file with keys : txt folder - result of tracking, dir with images, dir where to save visualisation, additional file (.segmap) located in mots_eval. example of run: python visualize_mots.py dir/with/txt /root/pointTrack/data/kittiRoot/testing/image_02 results /root/pointTrack/datasets/mots_tools/mots_eval/test.seqmap

EchoAmor commented 3 years ago

@Ilyabasharov Thanks again, I have found that and got the results . BTW, can u please share how to train the model on cars and pedestrians together? Now only got the car results, thanks a lot!

Ilyabasharov commented 3 years ago

@EchoAmor as I understood main problem is in spatial emb. it (posted code) supports only one class mb I am wrong. I solved it by train spatial emb on two classes but changed the annotation to one class. The second solution what I chose was to change spatial emb to other segmentation model witch can be faster

EchoAmor commented 3 years ago

@Ilyabasharov Thanks! I am working on spatial embeddding's code now , I thought about change other segmentation too, maybe later, thanks again!