Research map / Video / Action recognition
V4D: 4D Convolutional Neural Networks for Video-level Representation Learning
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 video-level 4D residual blocks to the 3D backbone | Mini-Kinetics, I3D-S ResNet18 · Mini-Kinetics | top-1 accuracy: 73.0 -> 75.6 | Helped |
| add video-level 4D residual blocks to the 3D backbone | Kinetics-400, ResNet50 · Kinetics-400 | top-1 accuracy: 74.9 -> 77.4 | Helped |
| add video-level 4D residual blocks to the 3D backbone | Something-Something-v1, ResNet50 · Something-Something-v1 | top-1 accuracy | Helped |
| use a 3×3×3×3 kernel instead of a 3×3×1×1 kernel (ablation) | V4D ResNet18 · Mini-Kinetics | top-1 accuracy: 74.5 -> 74.7 | Helped |
| use a 3×3×1×1 kernel instead of a 3×1×1×1 kernel (ablation) | V4D ResNet18 · Mini-Kinetics | top-1 accuracy: 73.8 -> 74.5 | Helped |
| add a 4D block at res3 alongside a block at res4 (ablation) | V4D ResNet18 · Mini-Kinetics | top-1 accuracy: 74.5 -> 75.6 | Helped |
| place the single 4D block at res5 rather than res4 (ablation) | V4D ResNet18 · Mini-Kinetics | top-1 accuracy: 74.5 -> 73.6 | Hurt |
| increase U_train from 3 to 4 (ablation) | V4D ResNet18 · Mini-Kinetics | top-1 accuracy: 74.3 -> 74.5 | Helped |
| increase U_train from 4 to 6 (ablation) | V4D ResNet18 · Mini-Kinetics | top-1 accuracy: 74.5 -> 74.6 | No clear effect |
| reverse the frames within each action unit (ablation) | V4D ResNet50 · Something-Something-v1 | top-1 accuracy: 50.4%→17.2% | Hurt |
| reverse the sequence of action units (ablation) | V4D ResNet50 · Something-Something-v1 | top-1 accuracy: 50.4%→20.1% | Hurt |
Closest papers by meaning
- 4D Spatio-Temporal ConvNets: Minkowski Convolutional Neural Networks (CVPR 2019 · similarity 0.89)
- Deep Analysis of CNN-Based Spatio-Temporal Representations for Action Recognition (CVPR 2021 · similarity 0.86)
- Geometry Guided Convolutional Neural Networks for Self-Supervised Video Representation Learning (CVPR 2018 · similarity 0.86)
- Can Spatiotemporal 3D CNNs Retrace the History of 2D CNNs and ImageNet? (CVPR 2018 · similarity 0.86)
- Uni4D: Unifying Visual Foundation Models for 4D Modeling from a Single Video (CVPR 2025 · similarity 0.84)
- Fused View-Time Attention and Feedforward Reconstruction for 4D Scene Generation (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.