Research map / Video / Video segmentation and tracking
LOMM: Latest Object Memory Management for Temporally Consistent Video Instance Segmentation
Seunghun Lee, Jiwan Seo, Minwoo Choi, Kiljoon Han, Jahoon Jeong, Zane Durante, Ehsan Adeli, Sang Hyun Park, Sunghoon Im
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 Latest Object Memory (LOM) (ablation) | DVIS baseline · YouTube-VIS 2022 (R50) | AP: 33.9 -> 39.2 | Improved |
| add Decoupled Object Association (DOA) (ablation) | DVIS baseline · YouTube-VIS 2022 (R50) | AP: 33.9 -> 36.2 | Improved |
| add DOA on top of LOM (ablation) | LOMM · YouTube-VIS 2022 (R50) | AP: 39.2 -> 41.1 | Improved |
| replace momentum memory with LOM (ablation) | MinVIS · YouTube-VIS 2022 (R50) | AP: 30.9 -> 34.0 | Improved |
| replace similarity memory with LOM (ablation) | MinVIS · YouTube-VIS 2022 (R50) | AP: 28.5 -> 34.0 | Improved |
| replace similarity/momentum memory with LOM (ablation) | DVIS · YouTube-VIS 2022 (R50) | AP: 35.8 -> 39.2 | Improved |
| add LOM memory (vs no memory) (ablation) | MinVIS · YouTube-VIS 2022 (R50) | AP: 26.8 -> 34.0 | Improved |
| refine aligned queries with tracker T_A (ablation) | LOMM · YouTube-VIS 2022 (R50) | AP: 39.1 -> 41.1 | Improved |
| use blended adaptive anchor instead of current-frame or memory anchor (ablation) | LOMM · YouTube-VIS 2022 (R50) | AP: 39.3 -> 41.1 | Improved |
| add occupancy guidance to Hungarian matching (ablation) | LOMM · YouTube-VIS 2022 (R50) | AP: 40.3 -> 41.1 | Improved |
| add early training strategy (ablation) | LOMM · YouTube-VIS 2022 (R50) | AP: 40.0 -> 41.1 | Improved |
| LOMM vs DVIS++ with ViT-L backbone | LOMM · YouTube-VIS 2019/2021/2022 | AP: +1.8 AP, +2.2 AP, +3.1 AP | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- DVIS++ — DVIS: Decoupled Video Instance Segmentation Framework (ICCV 2023) · better on 4 of 4 shared results
- CTVIS — CTVIS: Consistent Training for Online Video Instance Segmentation (ICCV 2023) · better on 1 of 1 shared result
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
- LVOS: A Benchmark for Long-term Video Object Segmentation (ICCV 2023 · similarity 0.88)
- LVOS: A Benchmark for Large-Scale Long-Term Video Object Segmentation (TPAMI 2025 · similarity 0.88)
- Target-Aware Object Discovery and Association for Unsupervised Video Multi-Object Segmentation (CVPR 2021 · similarity 0.88)
- Learning Position and Target Consistency for Memory-Based Video Object Segmentation (CVPR 2021 · similarity 0.87)
- Matching Anything by Segmenting Anything (CVPR 2024 · similarity 0.85)
- A Distractor-Aware Memory for Visual Object Tracking with SAM2 (CVPR 2025 · similarity 0.85)
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.
