PointNetGPD is an end-to-end grasp evaluation model to address the challenging problem of localizing robot grasp configurations directly from the point cloud.
Hi, Hongzhuo. above code have the chance filter all batch data, which will make batchnorm layer throw an erroe( valid batch_size <= 1).
besides increase the batchsize to decrease null valid bath probability. beside that, have you more elegant means to catch this case and skip this iteration?
https://github.com/lianghongzhuo/PointNetGPD/blob/336eb9e9cbb534dea36b3e92e0d1b7adef13d7f3/PointNetGPD/main_1v.py#L49C6-L49C6
Hi, Hongzhuo. above code have the chance filter all batch data, which will make batchnorm layer throw an erroe( valid batch_size <= 1). besides increase the batchsize to decrease null valid bath probability. beside that, have you more elegant means to catch this case and skip this iteration?