Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

add: example on semantic segmentation. #59

Merged
merged 2 commits into from
Feb 9, 2023
Merged

add: example on semantic segmentation. #59

merged 2 commits into from
Feb 9, 2023

Conversation

sayakpaul
Copy link
Member

Closes #45

There are two minor changes to the image classification example. I know this is not a good PR practice but the changes were extremely small.

@pacman100
Copy link
Contributor

pacman100 commented Feb 8, 2023

wrt semantic segmentation example:

  1. can we remove the cell ouputs of trainer.train() as they are huge and will take up considerable space
  2. I think the example notebook has bug because Accuracy Wall is nan for full finetuning too but in stevhliu/my_awesome_seg_model · Hugging Face it's very high. I tried full fine-tuning using 50 images as the task page (40 train + 10 eval) (https://huggingface.co/docs/transformers/tasks/semantic_segmentation#train)

@sayakpaul
Copy link
Member Author

I will address the first point in the follow-up commits. Regarding the second point, I think there are also quite a few nan values for the other classes in the repo you mentioned.

If you take a look at the training loss, it looks similar, though. I think the category that might be impacted because of the low number of training examples depends on their quantity in the training set as well. Not sure what you mean by the bug.

@pacman100
Copy link
Contributor

By bug I mean I'm unable to reproduce the same results for full fine-tuning using same train samples and eval samples along with same trainer arguments as that task page repo

@pacman100
Copy link
Contributor

@sayakpaul
Copy link
Member Author

@pacman100 addressed the things we discussed during the call. PTAL.

@pacman100
Copy link
Contributor

Thank you for iterating! LGTM 🤗

@sayakpaul sayakpaul merged commit 250edcc into huggingface:main Feb 9, 2023
@sayakpaul sayakpaul deleted the example/sem-seg branch February 9, 2023 06:39
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

Add an example demoing the use of PEFT for SegFormer Semantic Segmentation
2 participants