inference erro
Opened this issue · 2 comments
Hi, the pre-trained model is trained on 256 resolution. Therefore you need to resize the picture to 256. You can train the your own DQVAE on 512 resolution. We adopt the VQGAN's encoder and decoder structure to construct DQVAE. Therefore it should be easy to scale it to 512 resolution. Also, if you want to train the entropy-based DQVAE, you can firstly calculate the image entropy threshold by the script in scripts/tools/calculate_entropy_thresholds.py
.
Hi, the pre-trained model is trained on 256 resolution. Therefore you need to resize the picture to 256. You can train the your own DQVAE on 512 resolution. We adopt the VQGAN's encoder and decoder structure to construct DQVAE. Therefore it should be easy to scale it to 512 resolution. Also, if you want to train the entropy-based DQVAE, you can firstly calculate the image entropy threshold by the script in
scripts/tools/calculate_entropy_thresholds.py
.
Thank you for your answer. But I'm wondering, are you saying that inference an image at a certain resolution requires training the image at same resolution? I know that VQGAN can deduce images of arbitrary resolution under the condition of fixed sampling factor training