Research map / Video / Video segmentation and tracking
Recurrent Video Masked Autoencoders
Daniel Zoran, Nikhil Parthasarathy, Yi Yang, Drew A Hudson, João Carreira, Andrew Zisserman
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 more source frames (1 to 4) (ablation) | RVM-S · SSv2 | accuracy: 41.020 -> 52.34 | Improved |
| add more source frames (1 to 4) (ablation) | RVM-S · ScanNet depth | AbsRel: 1.596 -> 1.50 | Improved |
| add more source frames (1 to 4) (ablation) | RVM-S · Kinetics | accuracy: 39.280 -> 39.72 | No clear effect |
| use RNN aggregator instead of full self-attention (ablation) | RVM-S · SSv2 | accuracy: 49.1 -> 52.34 | Improved |
| use RNN aggregator instead of full self-attention (ablation) | RVM-S · ScanNet depth | AbsRel: 1.505 -> 1.50 | No clear effect |
| increase training examples 250M to 2B (ablation) | RVM-S · SSv2 | accuracy: 46.38 -> 57.20 | Improved |
| increase training examples 250M to 2B (ablation) | RVM-S · Kinetics | accuracy: 33.84 -> 47.70 | Improved |
| increase training examples 250M to 2B (ablation) | RVM-S · ScanNet depth | AbsRel: 1.75 -> 1.200 | Improved |
| replace image encoder with recurrent video encoder (RVM-L vs DINOv2-L) | RVM-L · Perception Test point tracking | AJ: 36.6 -> 77.3 | Improved |
| use RVM instead of VideoMAE for spatial correspondence | RVM-L · VIP | mIoU: 18.9 -> 38.0 | Improved |
| RVM-S versus distilled DINOv2-S on normalized average | RVM-S · Average over 8 tasks | normalized avg: 84.4 -> 96.1 | Improved |
| RVM-S versus distilled DINOv2-S on Kinetics | RVM-S · Kinetics | accuracy: 57.1 -> 49.6 | Hurt |
Closest papers by meaning
- VideoMAE V2: Scaling Video Masked Autoencoders With Dual Masking (CVPR 2023 · similarity 0.90)
- VideoMAC: Video Masked Autoencoders Meet ConvNets (CVPR 2024 · similarity 0.90)
- Extending Video Masked Autoencoders to 128 frames (NeurIPS 2024 · similarity 0.89)
- Video Language Model Pretraining with Spatio-temporal Masking (CVPR 2025 · similarity 0.88)
- EVEREST: Efficient Masked Video Autoencoder by Removing Redundant Spatiotemporal Tokens (ICML 2024 · similarity 0.88)
- Exploring Target Representations for Masked Autoencoders (ICLR 2024 · similarity 0.87)
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.
