Instructions to use zju-community/matchanything_eloftr with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use zju-community/matchanything_eloftr with Transformers:
# pip install -U transformers accelerate # Load model directly from transformers import AutoImageProcessor, AutoModelForKeypointMatching processor = AutoImageProcessor.from_pretrained("zju-community/matchanything_eloftr") model = AutoModelForKeypointMatching.from_pretrained("zju-community/matchanything_eloftr", device_map="auto") - Notebooks
- Google Colab
- Kaggle
ROMA safetensors
Will you make a safetensors version of the ROMA variant in your paper? I would like to try it and am unable to handle the zip file. Thank you!
Hi @alexdzm ,zju-community/matchanything_eloftr exists because EfficientLoFTR is implemented in HuggingFace Transformers library, which is not the case of RoMA.
If you wish to have RoMA also supported you can open an issue on the repo, I would probably take care of it myself or maybe someone from the community
Hi @stevenbucaille , thanks for your response. I understand! I have found a solution for my case.
FWIW its still of some use having the weights available as safetensors even if the model is not in transformers as some enterprise users cannot access .zips and .ckpt files easily.
There is already a thread about implementing RoMA https://github.com/huggingface/transformers/issues/36718 but it seems stale unfortunately.
Thanks for maintaining this repo!