microsoft
Failed to trigger correct segmentation model
I want to use the semantic segmentation model from this project. That‘s what I do to interact:
It is a strange result and I try again:
And for this time, I suspect that ChatGPT does not trigger the semantic segmentation model correctly.
I carefully read the related codes in visual_chatgpt.py and I am sure that the commands I entered should not cause ambiguity.
And I find out that the console give a correct action but choose a wrong model( image2pose????)
hi @ddzipp , can you provide these two images to me? I want to reproduce it to find out what went wrong. Thanks~
Sorry, the demo was running on Google's Colab and I didn't save the images. But I still came across similar examples after re-running the demo.
It seems that the segmentation model was triggered but it do not work very well.
What I find strange is that when the console shows Action as segmentation, the model chosen is Image2pose
hi @ddzipp, this is a typo that the processed function outputs should be modified to the correct Image2Seg, and it actually triggered the right Image2Seg model. Thanks very much! We will fix it.
Thank you for reply. I am a junior student who is interested in the direction of semantic segmentation. However, I found that Visual ChatGPT has poor semantic segmentation results for some specific scenarios. I hope I can explore further and make some small efforts. Thanks again for your reply and your job!