Making SEMat Hugging Face Transformers compatible

[NielsRogge]

Hi @XiaRho,

Congrats on your work. I discovered it from AK's daily papers: https://huggingface.co/papers/2410.06593. I opened https://huggingface.co/XiaRho/SEMat/discussions/1 to link the released models to it. Feel free to claim the paper as an author so that it appears at your HF account.

I was wondering if you're planning to make the model Transformers compatible, similar to the ViTMatte model? This would greatly increase its accessibility.

See here for a guide: https://huggingface.co/docs/transformers/custom_models. This would allow the following:

from transformers import SeMatForImageMatting

model = SeMatForImageMatting.from_pretrained("XiaRho/SEMat")

Besides that, are you planning to release the COCO-Matting dataset on the hub?

See here for a guide: https://huggingface.co/docs/datasets/image_dataset.

Let me know if you're interested/need any help.

Kind regards,

Niels from HF

[XiaRho]

Hi @NielsRogge,

Thank you very much for your kind words and for bringing my work to the attention of the Hugging Face community. I appreciate your suggestion to make the model Transformers compatible, which is indeed a great idea to enhance its accessibility. I'll consider working on this in my spare time.

As for the COCO-Matting dataset, I plan to first make it available via Google Drive for easier access. After that, I'll look into the possibility of uploading it to Hugging Face as well. I'm aware that having the dataset on the Hugging Face Hub would be very convenient for the community, and I'm open to exploring this option further.

Best regards,
XiaRho

[ygean]

Huggingface demo still raises error

[XiaRho]

@ygean Hi, we have fixed the error on the Huggingface online demo. However, due to the expiration of the PRO subscription on Huggingface, only the CPU can be used for inference, and a forward takes a few minutes. Therefore, I recommend using our gradio demo locally.

[ygean]

@XiaRho Thank you for your reply