How to train model on multiple classes?
RocketFlash opened this issue · 1 comments
RocketFlash commented
Hello, @zhixinwang !
I want to train model on multiple classes. I found that parameter num_classes not used in PointNetDet , and classification and regression outputs always same sized:
output_size = 3 + num_bins * 2 + NUM_SIZE_CLUSTER * 4
self.reg_out = nn.Conv1d(768, output_size, 1)
self.cls_out = nn.Conv1d(768, 2, 1)
How can I modify the outputs to train the network in 3 or more classes?
zhixinwang commented
Refer to #28. Close.