Research map / Video / Action recognition
What Can Simple Arithmetic Operations Do for Temporal Modeling?
Wenhao Wu, Yuxin Song, Zhun Sun, Jingdong Wang, Chang Xu, Wanli Ouyang
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 any single arithmetic operation (add/sub/mul/div) as temporal cue (ablation) | ResNet-18 temporal-irrespective baseline · Something-Something V1 | top-1 accuracy: 16.5% -> 46.5% (multiplication) | Improved |
| combine subtraction and multiplication (ablation) | ATM on ResNet-18 · Something-Something V1 | top-1 accuracy: 46.5% -> 48.2% | Improved |
| add addition on top of subtraction and multiplication (ablation) | ATM on ResNet-18 · Something-Something V1 | top-1 accuracy: 48.2% -> 48.1% | No clear effect |
| enlarge temporal context range Z from 1 to 4 (ablation) | ATM on ResNet-18 · Something-Something V1 | top-1 accuracy: 43.7% -> 48.2% | Improved |
| increase Z from 4 to 6 (ablation) | ATM on ResNet-18 · Something-Something V1 | top-1 accuracy: 48.2% -> 48.0% | No clear effect |
| add temporal convolution to ATM (ablation) | ATM on ResNet-18 · Something-Something V1 | top-1 accuracy: 48.2% -> 48.7% | Improved |
| add temporal convolution to ATM (ablation) | ATM on ViT-B/16 · Something-Something V1 | top-1 accuracy: 56.1% -> 58.1% | Improved |
| use conv stack feature extractor instead of FC or MLP (ablation) | ATM on ResNet-18 · Something-Something V1 | top-1 accuracy: 47.8% -> 48.7% | Improved |
| insert ATM at res5 instead of res3 (ablation) | ATM on ResNet-18 · Something-Something V1 | top-1 accuracy: 48.7% -> 43.1% | Hurt |
| insert ATM at res2 instead of res3 (ablation) | ATM on ResNet-18 · Something-Something V1 | top-1 accuracy: 48.7% -> 48.8% | No clear effect |
| insert ATM in middle ViT layer (7) instead of early layer (2) (ablation) | ATM on ViT-B/16 · Something-Something V1 | top-1 accuracy: 36.9% -> 58.1% | Improved |
| increase frames from 8 to 16 with ResNet-50 (ablation) | ATM · Something-Something V1 | top-1 accuracy: 53.9% -> 56.3% | Improved |
Closest papers by meaning
- Approximated Bilinear Modules for Temporal Modeling (ICCV 2019 · similarity 0.82)
- Enhancing Temporal Understanding in Video-LLMs through Stacked Temporal Attention in Vision Encoders (NeurIPS 2025 · similarity 0.81)
- Before It Fades: Reinforcing Temporal Representations at Inference Time in VideoLLMs (NeurIPS 2026 · similarity 0.81)
- TOMATO: Assessing Visual Temporal Reasoning Capabilities in Multimodal Foundation Models (ICLR 2025 · similarity 0.81)
- Temporal Reasoning Transfer from Text to Video (ICLR 2025 · similarity 0.81)
- MEET: Towards Memory-Efficient Temporal Sparse Deep Neural Networks (CVPR 2025 · similarity 0.81)
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.
