search

detectRecog / PointTrack

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

test on MOTSChallenge #7

Open cty-xiaobai opened 4 years ago

cty-xiaobai commented 4 years ago

hi,I trained a model for tracking people and verified it on the kitti dataset, but an error was reported on the MOTSchalleng dataset: RuntimeError: The size of tensor a (1080) must match the size of tensor b (1024) at non- singleton dimension 1

Looking forward to your reply!

detectRecog commented 3 years ago

When test on MOTSChallenge, you should adjust the 'Cluster' module (in utils/utils.py) to support the resolution of MOTSChallenge like:

` xm = torch.linspace(0, 2, 2048).view(1, 1, -1).expand(1, 2048, 2048)

ym = torch.linspace(0, 2, 2048).view(1, -1, 1).expand(1, 2048, 2048) `

EchoAmor commented 3 years ago

@cty-xiaobai Hi , can u share how to train on pedestrian and test on MOTS Challenge? I have no idea , but really need this , thanks very much!!!!