VAR Training and Inference Scripts

Hi, amazing work indeed !

I wonder, can you please release the training and inference scripts to be able to exactly reproduce your ImageNet 256x256 (and 512X512) results ?

In addition, can you please provide information about the number of nodes/GPUs used for each of the ImageNet256 and ImageNet512 experiments ?

Apr 04 '24 16:04 achen46

ty @achen46! You can check out https://github.com/FoundationVision/VAR/blob/main/demo_sample.ipynb for our inference demo. For training cost I'll post more details on readme soon. Our most expensive 512x512 experiment use 256xA100 for around 4days.

Apr 05 '24 00:04 keyu-tian

Thank you for your good work! I also wonder, can you release the inference scripts to be able to reproduce your results? I just tried to reproduce your result on ImageNet 256 x 256 using your checkpoint. Thank you very much.

Apr 05 '24 06:04 yyyouy

Hi @keyu-tian

Thank you so much for providing the pointer !

As mentioned, I am interested in a (shell) script that I can run to start training the model and be able to reproduce the results. Basically, all I need is a self-contained script (with all hyper-parameters) that can be run. Also obviously, need to know how many nodes are used.

256x256 ImageNet would suffice to begin with.

Apr 05 '24 15:04 achen46

any update?

Apr 13 '24 02:04 morestart

@achen46 @yyyouy @morestart I got swamped with work recently, but the training codes & scripts are now out! I'll post more scripts for inference and zero-shot demos in the next. Feel free to drop an issue again if you run into any problems.

Thanks, and have fun with VAR!

Apr 15 '24 22:04 keyu-tian