Research map / Video / Video segmentation and tracking
LV-MAE: Learning Long Video Representations through Masked-Embedding Autoencoders
Ilan Naiman, Emanuel Ben-Baruch, Oron Anschel, Alon Shoshan, Igor Kviatkovsky, Manoj Aggarwal, Gerard Medioni
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 LV-MAE masked-embedding pre-training over raw embeddings | LanguageBind embeddings · LVU avg | top-1 accuracy: 38.9 -> 63.4 | Improved |
| add LV-MAE pre-training over raw embeddings | LanguageBind embeddings · LVU Writer | top-1 accuracy: 5.95 -> 64.28 | Improved |
| add LV-MAE pre-training over raw embeddings | InternVideo2 embeddings · LVU avg | top-1 accuracy: 32.9 -> 59.6 | Improved |
| add MAE pre-training instead of end-to-end Mamba on clip embeddings (ablation) | LanguageBind embeddings · LVU avg | top-1 accuracy: 56.8 -> 63.4 | Improved |
| add MAE pre-training instead of end-to-end Transformer on clip embeddings (ablation) | LanguageBind embeddings · LVU avg | top-1 accuracy: 33.7 -> 63.4 | Improved |
| use LanguageBind clip embeddings instead of frame-MAE features (ablation) | LV-MAE · LVU avg | top-1 accuracy: 54.3 -> 63.4 | Improved |
| use attentive probing instead of linear probing (ablation) | LV-MAE LanguageBind · LVU avg | top-1 accuracy: 59.02 -> 63.41 | Improved |
| use semantic masking instead of random masking (ablation) | LV-MAE LanguageBind (attentive) · LVU avg | top-1 accuracy: 62.54 -> 63.41 | Improved |
| use semantic masking instead of random masking (ablation) | LV-MAE InternVideo2 (linear probing) · LVU avg | top-1 accuracy: 53.63 -> 52.60 | Hurt |
| increase encoder depth from 16 to 32 blocks (ablation) | LV-MAE · LVU avg | top-1 accuracy: 59.73 -> 63.41 | Improved |
| increase encoder depth from 24 to 32 blocks (ablation) | LV-MAE · COIN | top-1 accuracy: 93.26 -> 91.8 | Hurt |
| use 5 s segments instead of 10 s segments (ablation) | LV-MAE · LVU avg | top-1 accuracy: 61.43 -> 63.40 | 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.
- VideoMamba — VideoMamba: State Space Model for Efficient Video Understanding (ECCV 2024) · better on 2 of 3 shared results
- VideoBERT — VideoBERT: A Joint Model for Video and Language Representation Learning (ICCV 2019) · 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
- Extending Video Masked Autoencoders to 128 frames (NeurIPS 2024 · similarity 0.93)
- VideoMAC: Video Masked Autoencoders Meet ConvNets (CVPR 2024 · similarity 0.88)
- EVEREST: Efficient Masked Video Autoencoder by Removing Redundant Spatiotemporal Tokens (ICML 2024 · similarity 0.87)
- One Token per Highly Selective Frame: Towards Extreme Compression for Long Video Understanding (NeurIPS 2025 · similarity 0.87)
- VQToken: Neural Discrete Token Representation Learning for Extreme Token Reduction in Video Large Language Models (NeurIPS 2025 · similarity 0.87)
- Self-Guided Masked Autoencoder (NeurIPS 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.
