Thank you for your great work! I tried to train the RQ-VAE model on the FFHQ dataset with the default config configs/ffhq/stage1/ffhq256-rqvae-8x8x4.yaml. The batch size is 8 and I used one NVIDIA RTX-3090 GPU. However, I found the evaluation result is extremely worse (running it with the eval function from trainer_vqvae.py)
While in the model.train() mode the result is much better
So how to solve that question? Thank you a lot.
Thank you for your great work! I tried to train the RQ-VAE model on the FFHQ dataset with the default config
configs/ffhq/stage1/ffhq256-rqvae-8x8x4.yaml
. The batch size is 8 and I used one NVIDIA RTX-3090 GPU. However, I found the evaluation result is extremely worse (running it with theeval
function fromtrainer_vqvae.py
) While in themodel.train()
mode the result is much better So how to solve that question? Thank you a lot.