Getting same result for Random Generate

Question

Getting same result for Random Generate

cravisjan97 opened this issue 4 years ago · comments

I am trying to randomly generate a chair model. I ran the test_stacknewvae.py (with model.random_gen()) and ran GetOptimizedObj.m with use_struct=1 (use_struct=0 gave very bad results, don't know why). All 200 chair models were pretty similar (most of them were the same). How do I resolve this issue?

Jie Yang · Answer 1 · Tue Feb 09 2021 11:05:26 GMT+0800 (China Standard Time)

@cravisjan97 You should re-train the network with a higher weight of KL divergence. In the repo, we aim to get a better reconstruction to train the network with a small weight of KL divergence regularization.
If you have some questions, please feel free to contact us.

C Ravi Sundaram · Answer 2 · Wed Feb 10 2021 00:30:05 GMT+0800 (China Standard Time)

@tommaoer Thanks for the response. How can I change the KL-Divergence (is it l3 parameter) weight in train_stacknewvae.py? Are there any arguments that need to be set for running train_stacknewvae.py? Any suggestions for the other training hyperparameters?