Research map / Video / Video segmentation and tracking
MoNet: Deep Motion Exploitation for Video Object Segmentation
Huaxin Xiao, Jiashi Feng, Guosheng Lin, Yu Liu, Maojun Zhang
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 motion-aligned features from adjacent frames (ablation) | MoNet VGG16 DeepLab · DAVIS validation | Global mean: 75.7 -> 80.2 | Improved |
| Add the DT-based motion prior after feature alignment (ablation) | MoNet VGG16 DeepLab · DAVIS validation | Global mean: 80.2 -> 83.8 | Improved |
| Add fully-connected CRF post-processing (ablation) | MoNet with feature alignment and motion prior · DAVIS validation | Global mean: 83.8 -> 84.7 | Improved |
| Add feature alignment using both past and future neighbors instead of past only (ablation) | MoNet VGG16 DeepLab · DAVIS validation | Global mean: 79.9 -> 80.2 | Improved |
| Use conv5_3 rather than conv4_3 for feature alignment (ablation) | MoNet feature alignment · DAVIS validation | Global mean: 79.5 -> 80.2 | Improved |
| Increase the feature-alignment neighborhood from 3 to 5 frames (ablation) | MoNet conv5_3 feature alignment · DAVIS validation | Global mean: 80.2 -> 80.9 | Improved |
| Increase the feature-alignment neighborhood from 5 to 10 frames (ablation) | MoNet conv5_3 feature alignment · DAVIS validation | Global mean: 80.9 -> 79.6 | Hurt |
| Add motion-based alignment rather than directly combining neighboring features (ablation) | MoNet conv5_3 feature aggregation · DAVIS validation | Global mean: 77.4 -> 80.2 | Improved |
| Use the DT layer instead of MPNet to produce motion prior (ablation) | MoNet with feature alignment · DAVIS validation | J mean: 81.8 -> 81.6 | No clear effect |
| Use the DT layer instead of MPNet with LDOF (ablation) | Motion-prior extraction · DAVIS validation | J mean: 52.4 -> 50.8 | Hurt |
| Use the DT layer instead of MPNet with EpicFlow (ablation) | Motion-prior extraction · DAVIS validation | J mean: 56.9 -> 55.4 | Hurt |
| Use MoNet's motion exploitation rather than OSVOS | Video object segmentation · DAVIS validation | J mean: 79.8 -> 84.7 | Improved |
Later papers that compared with this method
1 later accepted paper reports MONet as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- SIMONe: View-Invariant, Temporally-Abstracted Object Representations via Unsupervised Video Decomposition (NeurIPS 2021 · better on 3 of 3 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
- Learning to Segment Moving Objects (IJCV 2018 · similarity 0.88)
- Learning Video Object Segmentation From Static Images (CVPR 2017 · similarity 0.88)
- A Survey on Deep Learning Technique for Video Segmentation (TPAMI 2022 · similarity 0.87)
- VONet: Unsupervised Video Object Learning With Parallel U-Net Attention and Object-wise Sequential VAE (ICLR 2024 · similarity 0.87)
- Segment Any Motion in Videos (CVPR 2025 · similarity 0.86)
- Learning Segmentation from Point Trajectories (NeurIPS 2024 · 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.
