Research map / Video / Video segmentation and tracking
Efficient-SAM2: Accelerating SAM2 with Object-Aware Visual Encoding and Memory Retrieval
Jing Zhang, Zhikai Li, Xuewen Liu, Qingyi Gu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add object-aware Sparse Window Routing to image encoder | SAM2.1-B+ · SA-V test | J&F: 77.7 -> 75.0 (1.69x encoder speedup) | Hurt |
| add Sparse Memory Retrieval to memory attention | SAM2.1-B+ · SA-V test | J&F: 77.7 -> 77.8 (1.82x) | No clear effect |
| use SMR saliency mask instead of random mask (ablation) | SMR · SA-V test (B+) | J&F: 76.7 -> 77.8 | Improved |
| use SMR saliency mask instead of uniform mask (ablation) | SMR · SA-V test (B+) | J&F: 76.8 -> 77.8 | Improved |
| add learned shortcut branch instead of identity bypass (ablation) | SWR · SA-V (SAM2.1-B+) | J&F: 73.0 -> 75.0 | Improved |
| use bottleneck FFN instead of full FFN shortcut (ablation) | SWR · SA-V (SAM2.1-L) | J&F: 79.2 -> 79.0 | No clear effect |
| add mask dilation to window selection (ablation) | SWR · SA-V test (B+) | J&F: 73.6 -> 75.0 | Improved |
| include all decoder predictions in window selection (ablation) | SWR · SA-V test (B+) | J&F: 73.6 -> 74.1 | Improved |
| raise memory sparsity beyond about 0.95 (ablation) | SMR · VOS accuracy | J&F | Hurt |
| combine SWR with SMR | Efficient-SAM2 · SA-V test (SAM2.1-B+, dt=1) | J&F: 77.7 -> 75.5 (1.74x) | Hurt |
| use token merging (ToMe) for image encoder instead of SWR | SAM2.1-B+ · SA-V test | J&F: 77.7 -> 55.3 (1.36x) | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- EdgeTAM — EdgeTAM: On-Device Track Anything Model (CVPR 2025) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SAM-I2V: Upgrading SAM to Support Promptable Video Segmentation with Less than 0.2% Training Cost (CVPR 2025 · similarity 0.90)
- SAM 2: Segment Anything in Images and Videos (ICLR 2025 · similarity 0.90)
- EfficientSAM: Leveraged Masked Image Pretraining for Efficient Segment Anything (CVPR 2024 · similarity 0.88)
- BA-SAM: Scalable Bias-Mode Attention Mask for Segment Anything Model (CVPR 2024 · similarity 0.87)
- A Distractor-Aware Memory for Visual Object Tracking with SAM2 (CVPR 2025 · similarity 0.87)
- FocSAM: Delving Deeply into Focused Objects in Segmenting Anything (CVPR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
