EdgeTAM for Core ML (Scut)

Meta's EdgeTAM, the on-device Track Anything Model (SAM 2 with a memory), converted to four compiled Core ML models. The Scut iOS video editor downloads them as an optional feature, Better cutout, which keeps the person you tap cut out through a clip.

Model What it does
EdgeTAMImageEncoder.mlmodelc encodes a frame
EdgeTAMInitializer.mlmodelc seeds the track from a tap or a box
EdgeTAMPropagator.mlmodelc carries the mask to the next frame from the memory
EdgeTAMMemoryEncoder.mlmodelc writes a frame's mask into the memory

Source

  • Code: facebookresearch/EdgeTAM at 7711e012a30a2402c4eaab637bdb00a521302c91
  • Checkpoint: facebook/EdgeTAM at 14d7ecc48c656b94e5184519f698cd5386c5a2bf, edgetam.pt, SHA-256 ed2d4850b8792c239689b043c47046ec239b6e808a3d9b6ae676c803fd8780df
  • Converted from Meta's own code and checkpoint by Scut's tools/edgetam/convert.py and packaged by tools/edgetam/package.sh. These are not a third party's converted binaries.

The app fetches the files at a pinned commit of this repository and checks each file's size and SHA-256 before using it.

Licence

Apache License 2.0, ยฉ Meta Platforms, Inc. and affiliates. The converted models keep the licence of the code and checkpoint they were made from.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for lkibbalam/scut-edgetam-coreml

Base model

facebook/EdgeTAM
Finetuned
(2)
this model