Open yydslhk opened 10 months ago
@yydslhk The psnr result of 33 has obviously not converged yet and you need to try training longer. I usually train 400K iteration with 128 x 128 patch size and 16 batch size, then 800k iteration with 256 x 256 patch size and 8 batch size.
To save time, you can continue to train based on the existing weights, usually I can reach 36+ psnr at tha stage of training with 128*128 patches @yydslhk
Hello, I obtained a PSNR of only 33 when training on the indoor dataset using your model. Can you please explain why?
Hi! I wonder how many GPUs and the batch size per GPU you used? @raindrop313
@yydslhk The psnr result of 33 has obviously not converged yet and you need to try training longer. I usually train 400K iteration with 128 x 128 patch size and 16 batch size, then 800k iteration with 256 x 256 patch size and 8 batch size.
hellow, May I ask if you can share your training time? I trained with 4090 GPU for 10 hours and only completed 65k iterations
Hello, I obtained a PSNR of only 33 when training on the indoor dataset using your model. Can you please explain why?