ggjy / delvm Goto Github PK

Hi, I am interested in this great job! Here I have a problem when running demo. I use LLaMA-1b-hf and vqgan-f16-8192-laion(As shown in data_preparition). The generated_img results:

(I find generated_img.max() = 0.07. Is there any mistake?
Thanks

Hi,

Thank you for your great work. I am facing the same problem as #2 (comment). Using the seg_1.png gives black and my inputid / output.sequence is the same as [Sutongtong233]'s. When I screenshot the seg_1.png, it works reasonably. Other images under the data folder are good as well.

I am using pytorch 1.13.1 and transformer 4.38.1. Could you give any advice to debug this? Or any insight that could cause this strange issue? Thank you and looking forward to your reply.

Hi

When you evaluate the model on image segmentation task, for calculating the accuracy, what post-processing did you use to align the prediction with categories? And can you please provide the evaluation code to calculate the ACC?

All the best,

恭喜作者做出这项不错的工作！
作者你好，请问这个工作是投的ICML2024吗？请问被录用了吗？审稿意见能否发一下呢？
谢谢，期待作者的回复。

Are you able to use this model to generate the next frame image directly? Also, do you happen to have any pre-trained models available?

how many A800 80G used for training？

Your profile photo are just like you! Niubility! I have been waiting LVM release code longlong time.

This work has a great performance on segment&pose&deraining. And did you test on more tasks? (Especially the 3D tasks such as
depth estimation, which the original LVM performance is good） In other words, I'm very curious about the multi-task capability of LVM model with less training data. Could u show more experimental results？

Hi, thanks for the great work.

The paper claims that you use an off-the-shelf VQGAN from Muse.

Could you kindly share which specific Muse code and pretrained model you used for this project?

Additionally, it would be really helpful to know where these resources can be found.

Thanks again!

Hi, thank you for releasing your great work!
I want to know the details of post-processing in foreground segmentation.
Can you provide some code about it? e.g., code on how to inference on PASCAL-5i.
Looking forward to your reply.