Research map / Video / Video segmentation and tracking
Learning to Segment Moving Objects
Pavel Tokmakov, Cordelia Schmid, Karteek Alahari
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 appearance stream (DeepLab fc6 features) (ablation) | video segmentation framework · DAVIS val | mean IoU: 43.5 -> 70.1 | Improved |
| add motion stream (MP-Net) (ablation) | video segmentation framework · DAVIS val | mean IoU: 59.6 -> 70.1 | Improved |
| add ConvGRU memory module (ablation) | video segmentation framework · DAVIS val | mean IoU: 64.1 -> 70.1 | Improved |
| use ConvGRU instead of ConvRNN (ablation) | memory module · DAVIS val | mean IoU: 68.7 -> 70.1 | Improved |
| use ConvGRU instead of ConvLSTM (ablation) | memory module · DAVIS val | mean IoU: 68.9 -> 70.1 | No clear effect |
| add bidirectional processing (ablation) | ConvGRU memory · DAVIS val | mean IoU: 67.2 -> 70.1 | Improved |
| use PASCAL VOC pretrained DeepLab instead of ImageNet-only (ablation) | appearance stream · DAVIS val | mean IoU: 64.1 -> 70.1 | Improved |
| replace DeepLab-v1 VGG16 with DeepLab-v2 ResNet101 (ablation) | appearance stream · DAVIS val | mean IoU: 70.1 -> 72.5 | Improved |
| use conv5 instead of fc6 features (ablation) | appearance stream · DAVIS val | mean IoU: 70.1 -> 67.7 | Hurt |
| train memory module on DAVIS instead of synthetic FT3D (ablation) | ConvGRU · DAVIS val | mean IoU: 55.3 -> 70.1 | Improved |
| use FlowNet 2.0 instead of LDOF flow for MP-Net (ablation) | MP-Net · DAVIS | mean IoU: 52.4 -> 62.6 | Improved |
| train MP-Net on FT3D + DAVIS instead of FT3D only (ablation) | MP-Net · DAVIS val | mean IoU: 62.6 -> 66.7 | Improved |
Closest papers by meaning
- Learning Video Object Segmentation With Visual Memory (ICCV 2017 · similarity 0.95)
- Learning Motion Patterns in Videos (CVPR 2017 · similarity 0.94)
- Appearance-based Refinement for Object-Centric Motion Segmentation (ECCV 2024 · similarity 0.91)
- Segmenting Moving Objects via an Object-Centric Layered Representation (NeurIPS 2022 · similarity 0.91)
- Learning Segmentation from Point Trajectories (NeurIPS 2024 · similarity 0.91)
- Segment Any Motion in Videos (CVPR 2025 · similarity 0.90)
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.
