Research map / Video / Video segmentation and tracking
CT-Net: Channel Tensorization Network for Video Classification
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 |
|---|---|---|---|
| replace the ResNet convolution with CT-Module (ablation) | ResNet-50 · Something-Something V1 | Top-1: 16.9 vs. 47.3 | Helped |
| use 2 channel sub-dimensions rather than 1 (ablation) | CT-Module · Something-Something V1 | Top-1: 46.5 -> 47.3 | Helped |
| increase tensorization from 2 to 4 sub-dimensions (ablation) | CT-Module · Something-Something V1 | Top-1: 47.3 -> 46.5 | Hurt |
| use parallel rather than serial spatial-temporal convolution (ablation) | CT-Module · Something-Something V1 | Top-1: 47.2 -> 47.3 | Helped |
| set C2 to the rounded middle size ⌊√C⌋ (ablation) | CT-Module · Something-Something V1 | Top-1: 47.0, 46.8, 47.2 -> 47.3 | Helped |
| add point-wise convolution (ablation) | CT-Module · Something-Something V1 | Top-1: 47.3 -> 48.0 | Helped |
| add Tensor Excitation rather than SE (ablation) | CT-Module + PWConv · Something-Something V1 | Top-1: 48.8 -> 50.1 | Helped |
| increase the spatial-temporal kernel size (ablation) | CT-Module · Something-Something V1 | Top-1: 47.3 -> 47.6 | Helped |
| replace 7 blocks rather than 6 with CT-Blocks (ablation) | ResNet-50 · Something-Something V1 | Top-1: 47.2 -> 47.3 | Helped |
| train and test at 256×256 rather than train at 224×224 and test at 256×256 (ablation) | CT-Net · Something-Something V1 | Top-1: 49.7 -> 50.1 | Helped |
| use 24 input frames instead of 16 (ablation) | CT-Net · Something-Something V1 | Top-1: 52.5 -> 52.5 | No clear effect |
| use a CT-Net ensemble with multiple frame counts (ablation) | CT-Net · Something-Something V1 | Top-1: 52.5 -> 56.6 | Helped |
Closest papers by meaning
- T-Net: Parametrizing Fully Convolutional Nets With a Single High-Order Tensor (CVPR 2019 · similarity 0.88)
- Convolutions and More as Einsum: A Tensor Network Perspective with Advances for Second-Order Methods (NeurIPS 2024 · similarity 0.88)
- ChannelNets: Compact and Efficient Convolutional Neural Networks via Channel-Wise Convolutions (NeurIPS 2018 · similarity 0.87)
- Exploring Unexplored Tensor Network Decompositions for Convolutional Neural Networks (NeurIPS 2019 · similarity 0.86)
- Tensor-Train Recurrent Neural Networks for Video Classification (ICML 2017 · similarity 0.86)
- Coarse-To-Fine Tensor Trains for Compact Visual Representations (ICML 2024 · 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.