devzhk / maskdit Goto Github PK

Thank you for your Nice work!
Is there any possible to use different sampling algorithm for generating, like DPMS, DDPM etc.

Nice work!
I've been training MaskDiT on my own dataset recently, but I'm unsure if the loss is decreasing normally because I'm not seeing satisfactory results in the generated images. Can you please provide the log records of training on ImageNet 256x256? Or the loss curve.

Thanks very much!

Hi,

I wanted to try out your work, however, I have a forbidden access when trying to download the checkpoints using the provided links.

wget https://maskdit-bucket.s3.us-west-2.amazonaws.com/cos-2037500.pthttps://maskdit-bucket.s3.us-west-2.amazonaws.com/imagenet256-ckpt-best_without_guidance.pt

--2023-06-19 19:51:10--  https://maskdit-bucket.s3.us-west-2.amazonaws.com/cos-2037500.pthttps://maskdit-bucket.s3.us-west-2.amazonaws.com/imagenet256-ckpt-best_without_guidance.pt
Resolving maskdit-bucket.s3.us-west-2.amazonaws.com (maskdit-bucket.s3.us-west-2.amazonaws.com)... 52.92.227.154, 52.92.196.154, 52.218.242.129, ...
Connecting to maskdit-bucket.s3.us-west-2.amazonaws.com (maskdit-bucket.s3.us-west-2.amazonaws.com)|52.92.227.154|:443... connected.
HTTP request sent, awaiting response... 403 Forbidden
2023-06-19 19:51:11 ERROR 403: Forbidden.

Best,

I test MaskDiT and DiT training on LSUNChurch dataset. But only find the training speed almost the same, but FID much worse.
Both set num_classes to 1, to imitate unconditional training. Any idea about this?

I download a .json file from https://github.com/anishathalye/imagenet-simple-labels/blob/master/imagenet-simple-labels.json, which also consists of 1,000 classes. Is your imagenet_label.json in generate.py the same?

Do u use the extract_imagenet_feature.py in the U-ViT to get latent ImageNet dataset?