MoVQGAN

MoVQGAN copied to clipboard

About more detailed configuration for model reproduction

1

Awesome work! However, I somehow encountered the difficulty reproducing the performance of the autoencoder by training-from-scratch. Especially with the addition of GAN loss, getting collapsed reconstruction results. I used a...

minkowski0125

Is perceptual loss using a random initialized VGG?

3

Hi, thanks for sharing the code. [Here](https://github.com/ai-forever/MoVQGAN/blob/main/movqgan/modules/losses/lpips.py#L17) seems that you are using a `pretrained=False` VGG network. I wonder if this is the common practice? Thanks

bladewaltz1

I modified the 67M version configs to train the network with 384\*384 imgs, the size of z changed to 48\*48\*4. However it shown "cuda out of memory", even on an...

Droliven

Admiring work! Do you have a fp16 version which would be better for other works like text-to-image?

tengjiayan20

Hello, may I ask you how does the hyperparameters n_embed and embed_dim influence the performance, could you give me some insight

cannonli7

MOVQ, decode_code: use grid_size and batch_size variables in reshape

1

thanks for this repo, its great. Can we get this upstreamed so its in the pip package?

iejMac

I want to training from a small size image like cifar-10, wondering how to do it.

lagokun

Could you please share the basic configuration file of Mo-VQGAN?

UESTC-Med424-JYX