Research map / Video / Action recognition
MixANT: Observation-dependent Memory Propagation for Stochastic Dense Action Anticipation
Syed Talal Wasim, Hamid Suleman, Olga Zatsarynna, Muzammal Naseer, Juergen Gall
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 observation-dependent mixture of A matrices (MixANT) versus MANTA | MANTA · Breakfast (alpha=0.2, beta=0.1) | Mean MoC: 27.7 -> 29.6 | Improved |
| add mixture of A matrices (MixANT) versus MANTA | MANTA · Assembly101 (alpha=0.2, beta=0.1) | Top-1 MoC: 16.9 -> 20.3 | Improved |
| scale experts from 1 to about 5 per block (ablation) | MixANT · Breakfast (alpha=0.2, beta=0.1) | Top-1 MoC | Improved |
| increase experts beyond 5 (ablation) | MixANT · Breakfast (alpha=0.2, beta=0.1) | Top-1 MoC | Hurt |
| add initial static Mamba blocks up to K0=3 (ablation) | MixANT · Breakfast (alpha=0.2, beta=0.1) | Top-1 MoC: 52.4 -> 53.5 | Improved |
| use unified router for forward and backward SSM instead of independent routers (ablation) | MixANT · Breakfast (alpha=0.2, beta=0.1) | Top-1 MoC | Improved |
| add load balancing loss (ablation) | MixANT · Breakfast (alpha=0.2, beta=0.1) | Top-1 MoC | Improved |
| raise load balancing coefficient beyond 0.15 (ablation) | MixANT · Breakfast (alpha=0.2, beta=0.1) | Top-1 MoC | Hurt |
| condition gating vector only on observed frames rather than observed plus zero-padded future (ablation) | MixANT · Breakfast | Top-1 MoC | Improved |
| use mixture of experts instead of query-key for input-dependent A (ablation) | MixANT · Breakfast Top-1 MoC averaged over all alpha/beta | Top-1 MoC: 52.3 -> 54.1 | Improved |
| use mixture of experts instead of large MLP for input-dependent A (ablation) | MixANT · Breakfast Top-1 MoC averaged over all alpha/beta | Top-1 MoC: 47.4 -> 54.1 | Improved |
| replace Mamba block with MixMamba in Testra | Testra short Mamba · EK-100 overall | action recall@5: 15.2 -> 17.1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MANTA — Manta: Enhancing Mamba for Few-Shot Action Recognition of Long Sub-Sequence (AAAI 2025) · better on 13 of 14 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
- MANTA: Diffusion Mamba for Efficient and Effective Stochastic Long-Term Dense Action Anticipation (CVPR 2025 · similarity 0.87)
- Predictive Attractor Models (NeurIPS 2024 · similarity 0.85)
- AntGPT: Can Large Language Models Help Long-term Action Anticipation from Videos? (ICLR 2024 · similarity 0.85)
- StateSpaceDiffuser: Bringing Long Context to Diffusion World Models (NeurIPS 2025 · similarity 0.85)
- BLEND: Behavior-guided Neural Population Dynamics Modeling via Privileged Knowledge Distillation (ICLR 2025 · similarity 0.85)
- Quantifying Memory Utilization with Effective State-Size (ICML 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.
