search

ethnhe / PVN3D

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

training on Linemod #98

Open shockjiang opened 2 years ago

shockjiang commented 2 years ago

Hi, I try to reproduce your experiments on Linemod. Since you have different models for different objects, does that imply you split dataset to multiple and train for different objects?

e.g., there are 10000 records in total, and 1000 records for ape. So do you train the ape mode on this 1000 subset?

shockjiang commented 2 years ago

what's more, acc-0.1d for nearly 100 percents. That is really amazing. Since diameter of those objects are around 10-20cm, that means almost all add are in 1-2cm. Really impressive.