search

facebookresearch / segment-anything

The repository provides code for running inference with the SegmentAnything Model (SAM), links for downloading the trained model checkpoints, and example notebooks that show how to use the model.
Apache License 2.0
47.42k stars 5.61k forks source link

Why not provide transfer results on object detection task? #546

Open QingSuML opened 1 year ago

QingSuML commented 1 year ago

The paper provides results only in segmentation domains. As it is model targeting dense prediction task, it would be interesting to see its transfer performance on other dense prediction tasks, such as object detection and depth prediction.

Would it be possible to obtain the results pertaining to object detection and depth prediction from your study? This information would establish a constructive baseline for forthcoming research endeavors, enabling meaningful comparisons.

Thanks

heyoeyo commented 1 year ago

Since SAM uses a vit model to handle image encoding, it arguably already is a 'transfer result' in a sense (i.e it's an example of how the vit models do on segmentation tasks).

That being said, there is a neat repo: awesome-segment-anything which someone has put together showing SAM projects applied to different use cases.