search

ethnhe / PVN3D

Code for "PVN3D: A Deep Point-wise 3D Keypoints Hough Voting Network for 6DoF Pose Estimation", CVPR 2020
MIT License
488 stars 105 forks source link

Why big loss? #53

Closed Ixion46 closed 3 years ago

Ixion46 commented 3 years ago

Hey @ethnhe I hope you can help me again. I have created my own dataset in the structure of ycb and started to train it. My depth images are created in NDDS and are in cm, so i took factor_depth = 100, is this correct? What can be another reason of this high loss and what would be a "good" result? loss

ethnhe commented 3 years ago

Check that the final output point cloud in your dataset scripts is in m. Visualize and check that your train and validate set is correct by running your dataset scripts, for example, python3 -m datasets.ycb.ycb_dataset for YCB-Video dataset.

ethnhe commented 3 years ago

The guide for adaptation to new datasets has been added in README.