qq456cvb/KeypointNet

Question about training strategy

Closed this issue · 2 comments

slcheng97 commented

As for the training process on the KeypointNet dataset, should I train my model for each class one by one? Or can I train it on the whole dataset?

qq456cvb commented

Currently, our released code trains the model for each class one by one considering intra-class consistency. However, it should be straightforward to train a universal network (by setting number of keypoints to the maximum among classes).

  • 👍1
slcheng97 commented

Got it, thanks a lot!