Integrate EfficientTAM

[cjaverliat]

Integrate EfficientTAM

This PR integrates the EfficientTAM model from the paper "Efficient Track Anything". The components originate from the repository of the authors and the config file were slightly adapted for being usable in this repository.

"Efficient Track Anything"
Yunyang Xiong, Chong Zhou, Xiaoyu Xiang, Lemeng Wu, Chenchen Zhu, Zechun Liu, Saksham Suri, Balakrishnan Varadarajan, Ramya Akula, Forrest Iandola, Raghuraman Krishnamoorthi, Bilge Soran, Vikas Chandra
arXiv:2411.18933 (Paper Link)

Usage

Download the weights by running

cd checkpoints
./download_ckpts_efficienttam.sh

The models can then be loaded as any other models (the video predictor notebook was updated for an example):

from sam2.build_sam import build_sam2_video_predictor

sam2_checkpoint = "../checkpoints/efficienttam_s_1.pt"
model_cfg = "configs/efficienttam/efficienttam_s_1.yaml"

predictor = build_sam2_video_predictor(model_cfg, sam2_checkpoint, device=device)