Research map / Video / Video segmentation and tracking
VA-RED$^2$: Video Adaptive Redundancy Reduction
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 input-dependent temporal and channel redundancy reduction | R(2+1)D-18 · Mini-Kinetics-200 | clip-1 accuracy: 57.5 -> 60.6 | Helped |
| Add input-dependent temporal and channel redundancy reduction | I3D-InceptionV2 · Mini-Kinetics-200 | GFLOPs: 56.0 -> 26.5 | Helped |
| Add input-dependent temporal and channel redundancy reduction | X3D-M · Mini-Kinetics-200 | clip-1 accuracy: 66.5 -> 65.5 | Hurt |
| Add input-dependent temporal and channel redundancy reduction | R(2+1)D-18 · Kinetics-400, 16-frame | clip-1 accuracy: 57.3 -> 58.4 | Helped |
| Add channel-wise redundancy reduction | Dilated ResNet-18 · ADE20K validation | mean IoU: 31.2\% -> 31.3\% | Helped |
| Disable dynamic channel modeling (retain temporal modeling) (ablation) | VA-RED² R(2+1)D-18 · Mini-Kinetics-200, 8-frame | clip-1 accuracy: 57.7 -> 57.1 | Hurt |
| Disable dynamic temporal modeling (retain channel modeling) (ablation) | VA-RED² R(2+1)D-18 · Mini-Kinetics-200, 8-frame | clip-1 accuracy: 57.7 -> 57.0 | Hurt |
| Disable dynamic channel modeling (retain temporal modeling) (ablation) | VA-RED² R(2+1)D-18 · Mini-Kinetics-200, 16-frame | GFLOPs: 40.4 -> 46.1 | Hurt |
| Disable dynamic temporal modeling (retain channel modeling) (ablation) | VA-RED² R(2+1)D-18 · Mini-Kinetics-200, 16-frame | GFLOPs: 40.4 -> 46.3 | Hurt |
| Add efficiency-loss finetuning (ablation) | VA-RED² R(2+1)D-18 · Kinetics-400 | GFLOPs: 49.8 -> 40.3 | Helped |
| Add efficiency-loss finetuning (ablation) | VA-RED² R(2+1)D-18 · Kinetics-400 | clip-1 accuracy: 57.9 -> 58.4 | Helped |
| Use search space 3 instead of search space 2 (ablation) | VA-RED² R(2+1)D-18 · Mini-Kinetics-200, 8-frame | clip-1 accuracy: 58.2 vs 57.7 | Mixed |
Closest papers by meaning
- FastVID: Dynamic Density Pruning for Fast Video Large Language Models (NeurIPS 2025 · similarity 0.85)
- EchoPrune: Interpreting Redundancy as Temporal Echoes for Efficient VideoLLMs (NeurIPS 2026 · similarity 0.84)
- Scaling Video-Language Models to 10K Frames via Hierarchical Differential Distillation (ICML 2025 · similarity 0.84)
- FlowCut: Rethinking Redundancy via Information Flow for Efficient Vision-Language Models (NeurIPS 2025 · similarity 0.84)
- VideoLLM-MoD: Efficient Video-Language Streaming with Mixture-of-Depths Vision Computation (NeurIPS 2024 · similarity 0.84)
- Instant Video Models: Universal Adapters for Stabilizing Image-Based Networks (NeurIPS 2025 · similarity 0.84)
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.