Research map / Video / Video segmentation and tracking
MobileVOS: Real-Time Video Object Segmentation Contrastive Learning Meets Knowledge Distillation
Roy Miles, Mehmet Kerim Yucel, Bruno Manganelli, Albert Saà-Garriga
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 logit distillation to cross-entropy (ablation) | MobileVOS student · YouTube-VOS validation | J&F: 80.75 -> 81.38 | Improved |
| Add representation distillation alongside logit distillation (ablation) | MobileVOS student · YouTube-VOS validation | J&F: 81.38 -> 82.28 | Improved |
| Use representation distillation rather than the ω=0 supervised-contrastive endpoint (ablation) | MobileVOS ResNet18 · DAVIS 2017 validation | J&F: 83.1 -> 84.2 | Improved |
| Use supervised contrastive learning rather than the ω=1 distillation endpoint (ablation) | MobileVOS ResNet18 · YouTube-VOS validation | J&F: 81.1 -> 82.3 | Improved |
| Add distillation (ablation) | MobileVOS MobileNetV2 without ASPP · DAVIS 2016 validation | J&F: 89.2 -> 90.1 | Improved |
| Add distillation (ablation) | MobileVOS MobileNetV2 without ASPP · DAVIS 2017 validation | J&F: 80.5 -> 81.8 | Improved |
| Remove ASPP (ablation) | MobileVOS MobileNetV2 · DAVIS 2016 validation | J&F: 90.5 -> 90.1 | Hurt |
| Remove ASPP (ablation) | MobileVOS MobileNetV2 · DAVIS 2017 validation | J&F: 82.2 -> 81.8 | Hurt |
| Remove ASPP (ablation) | MobileVOS MobileNetV2 · YouTube-VOS 2019 validation | Overall: 80.3 -> 80.1 | Mixed |
| Add poly loss (ablation) | MobileNetV2 without ASPP, trained without distillation · DAVIS 2016 validation | J&F: 89.8 -> 89.8 | No clear effect |
| Add poly loss (ablation) | MobileNetV2 without ASPP, trained without distillation · DAVIS 2017 validation | J&F: 79.9 -> 80.1 | No clear effect |
| Use boundary sampling instead of random pixel sampling (ablation) | MobileVOS representation distillation · DAVIS 2016 validation | J&F | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 2 on most shared results.
- XMem†* — XMem: Long-Term Video Object Segmentation with an Atkinson-Shiffrin Memory Model (ECCV 2022) · better on 0 of 3 shared results
- BATMAN — BATMAN: Bilateral Attention Transformer in Motion-Appearance Neighboring Space for Video Object Segmentation (ECCV 2022) · better on 0 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
- Revisiting Contrastive Methods for Unsupervised Learning of Visual Representations (NeurIPS 2021 · similarity 0.87)
- Space-Time Memory Networks for Video Object Segmentation With User Guidance (TPAMI 2020 · similarity 0.87)
- Augmentation-Free Dense Contrastive Knowledge Distillation for Efficient Semantic Segmentation (NeurIPS 2023 · similarity 0.87)
- MobileSAM2: Lightweight Segment Anything in Images and Videos via Hypergraphical Knowledge Distillation (ECCV 2026 · similarity 0.87)
- LiVOS: Light Video Object Segmentation with Gated Linear Matching (CVPR 2025 · similarity 0.86)
- Learning Spatial-Semantic Features for Robust Video Object Segmentation (ICLR 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.
