search

NVlabs / ODISE

Official PyTorch implementation of ODISE: Open-Vocabulary Panoptic Segmentation with Text-to-Image Diffusion Models [CVPR 2023 Highlight]
https://arxiv.org/abs/2303.04803
Other
862 stars 49 forks source link

error in demo #2

Open yangydeng opened 1 year ago

yangydeng commented 1 year ago

Hello, GREAT JOB! It looks like your demo shows an error when inference, do you have any plan to fix it? Looking forward to play with it :D

RogerQi commented 1 year ago

It works on my end. Sometimes Huggingface throws an error and resubmitting usually works.

cipri-tom commented 1 year ago

Are you referring to the einops related error ? AttributeError: module 'keras.backend' has no attribute 'is_tensor' . It seems to stem from einops==0.3.0 in sdkit . I've solved this with pip install --upgrade einops then restarting the kernel.

If you get an error related to RGBA values, then you just need to run again. The colors are generated randomly, so maybe sometimes the value is not good for being a color

yan-xu-helm commented 1 year ago

Are you sure the RGBA error is due to randomness? I ran it multiple times and got the error every single time

rmpku commented 1 year ago

Are you referring to the einops related error ? AttributeError: module 'keras.backend' has no attribute 'is_tensor' . It seems to stem from einops==0.3.0 in sdkit . I've solved this with pip install --upgrade einops then restarting the kernel.

If you get an error related to RGBA values, then you just need to run again. The colors are generated randomly, so maybe sometimes the value is not good for being a color

It's work for me, thanks!