search

bytedance / IRASim

Apache License 2.0
76 stars 4 forks source link

How to run the `languagetable.py` #2

Closed Lr-2002 closed 2 months ago

Lr-2002 commented 3 months ago

Hi, thanx for your impressing work. I'm trying to use the languagetable.py to make sure how to work with IRASim While Now the 'args.base_dir' is missing and It seems that the vae model is not involved in the pretrained_models, Could you give more advice in re-run the languagetable.py ?

fangqi-Zhu commented 2 months ago

Hi, thank you for your attention.

You can set args.base_dir = '' on line 62 of util.py. I have also updated the code to include this line. args.base_dir is a common prefix for data storage, intended for switching between different cloud drives.

You can download the VAE model (stabilityai/stable-diffusion-xl-base-1.0) from Hugging Face.

Lr-2002 commented 2 months ago

okk, Thanks a lot. Is it necessary to dowload all the train, eval, ckpt to run the languagetable.py? I've just download eval and ckpt, It seems not easy to run the code.

fangqi-Zhu commented 2 months ago

To run application/languagetable.py, you just need to download the checkpoint. I have checked the code before publishing; just some paths need to be set. Feel free to ask if you encounter any issues with the code.

Lr-2002 commented 2 months ago

ok, could I have your wechat to talk about the detail in re-produce? Maybe i could give more detail about re-product for this great work after run it from scratch!

Lr-2002 commented 2 months ago

What's more the languagetable.py seems to relay on latent video and video path a lot, It might be some mistake in pushing?

Lr-2002 commented 2 months ago

ok.. I've fixed all the problem. But to be honest now the code seems have some problem for only checkpoints.tar.gz download

Lr-2002 commented 2 months ago

to be honest... you push the previous-commit absolutly... the language table has too much not-related code

fangqi-Zhu commented 2 months ago

Sorry, I forgot that language_table.py depends on the eval dataset. Using the eval dataset, you can try more examples, or you need to modify the code to encode the given image provided on GitHub. And I believe there is no unrelated code in language_table.py since code is cleaned before publishing. Only paths you need to setup to run the code. Feel free to reach out if you encounter other bugs.

Lr-2002 commented 2 months ago

Okay, Thx a lot. There is no more problem and I'l close the Issue. Thanks for your work and patient help