search

NVlabs / PoseCNN-PyTorch

PyTorch implementation of the PoseCNN framework
Other
195 stars 49 forks source link

Multiple objects of the same class #7

Closed hrishikeshkale4 closed 3 years ago

hrishikeshkale4 commented 3 years ago

I am trying to train this network for a new dataset which has images having multiple objects of the same class. When I went through this code I found that the arrays 'gt_boxes' and 'pose_blob' that is computed in the data loader (and later passed on to the cuda layers) have size num_classes x 5 and num_classes x 9 respectively. As per my understanding this representation assumes that there is going to be atmost one object of each class in any image. Could you please let me know how can I modify this to handle multiple objects of the same class ?

Any help in this regard is highly appreciated

Thanks Hrishikesh

yuxng commented 3 years ago

Hough voting supports detection of multiple object instances of the same class. During training, we only train with an object instance per class, but it can be applied to detect multiple instances in testing.