search

zju3dv / clean-pvnet

Code for "PVNet: Pixel-wise Voting Network for 6DoF Pose Estimation" CVPR 2019 oral
Apache License 2.0
400 stars 106 forks source link

Questions about Two-Class or more Setting #265

Closed Divyam10 closed 1 year ago

Divyam10 commented 1 year ago

Hello, thanks for publishing code to this amazing paper!

  1. I am trying to the train the data for two classes as an experiment. I achieved good results by training them with Ver_Dim : 18 and Seg_Dim: 3. Even though it is multi-class it detects only one object at a time, obviously. I read the issue #57 and it that you mentioned we will have to change ransac_voting_layer_v3 to make it generate keypoints for two classes. Can you throw some light on how to do that?
  2. As a second experiment, I am trying to train two classes using ver_dim : 36 and Seg_dim : 3. In this the vertex data for class not present in the image should be zeroes? As my data consists of only one class per image.
pengsida commented 1 year ago

To detect more classes of object, an easier way is to add an object detector before PVNet, which is implemented in this code.