search

chensong1995 / HybridPose

HybridPose: 6D Object Pose Estimation under Hybrid Representation (CVPR 2020)
MIT License
412 stars 64 forks source link

Confusion in using Pretrained weights. #30

Open vikas-baghel opened 4 years ago

vikas-baghel commented 4 years ago

Hi, I wanted to see how the model is working so I thought I would download the pre-trained weights and give some images as input and see the 3D bounding box as output. But I can see that there are different weights for different classes. Why there are different weights? I am new in this field. As far as I know, we require only one pre-trained weight for all the classes? Please help me with this. Sorry if I am missing some obvious things.

chensong1995 commented 4 years ago

Hello Vikas,

Thanks for you interest in our research!

We follow the convention in PVNet and train a new network for every object class. However, it is easy to modify the model and train one single network for all object classes. You will need to change the network output to predict multi-class semantic segmentation rather than binary semantic segmentation masks. You can then extract intermediate representations from each object in the image, and feed them to the pose regression module described in the paper.

I hope this helps, Chen