search

franktpmvu / NeighborTrack

[CVPR 2023 workshop] NeighborTrack: Single Object Tracking by Bipartite Matching With Neighbor Tracklets and Its Applications to Sports
46 stars 1 forks source link

vot challenge code #10

Open aqizhoua opened 3 weeks ago

aqizhoua commented 3 weeks ago

Thank you for your excellent work! Is your code for training on the vot challenge open source? I would like to refer to it, thank you!

franktpmvu commented 3 weeks ago

hi, we use the pre-train model from OStrack, because this method is a post-processing method, so this method doesn't need training, you can download the OStrack model and then run our code.

aqizhoua commented 3 weeks ago

Thank you for your quick response! Where is the code for OSTrack applied to VOT? I have used it to train on got10k, coco, trackingnet and lasot, but your code should have been improved on vot, such as generating masks? I want to ask about this code. Thank you!

franktpmvu commented 3 weeks ago

We are involved in the VOT2022 bbox challenge, so in that time we use the original OStrack model (output bounding box), but you can change the base model with segmentation output to join the semantic segmentation tracking challenges. the main condition is that the model can output neighbors (which means not just get only one tracking segmentation output, you need the candidates with a confidence score).

aqizhoua commented 3 weeks ago

Thank you for your quick and patient reply! I will give it a try, thank you very much!