Research map / Video / Action recognition
Learning Spatio-Temporal Representation With Pseudo-3D Residual Networks
Zhaofan Qiu, Ting Yao, Tao Mei
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 |
|---|---|---|---|
| replace 2D residual units by P3D-A blocks | ResNet-50 · UCF101 split1 (fine-tuned from ResNet-50) | accuracy (%): 80.8% -> 83.7% | Improved |
| replace 2D residual units by P3D-B blocks | ResNet-50 · UCF101 split1 (fine-tuned from ResNet-50) | accuracy (%): 80.8% -> 82.8% | Improved |
| replace 2D residual units by P3D-C blocks | ResNet-50 · UCF101 split1 (fine-tuned from ResNet-50) | accuracy (%): 80.8% -> 83.0% | Improved |
| mix P3D-A, P3D-B and P3D-C blocks instead of only P3D-A (ablation) | P3D ResNet · UCF101 split1 (fine-tuned from ResNet-50) | accuracy (%): 83.7% -> 84.2% | Improved |
| mix P3D-A, P3D-B and P3D-C blocks instead of only P3D-B (ablation) | P3D ResNet · UCF101 split1 (fine-tuned from ResNet-50) | accuracy (%): 82.8% -> 84.2% | Improved |
| mix P3D-A, P3D-B and P3D-C blocks instead of only P3D-C (ablation) | P3D ResNet · UCF101 split1 (fine-tuned from ResNet-50) | accuracy (%): 83.0% -> 84.2% | Improved |
| add temporal 1D convolutions (P3D) to ResNet-152 | ResNet-152 · Sports-1M | video hit@1 (%): 64.6% -> 66.4% | Improved |
| use decomposed 2D+1D convolutions instead of 3D convolutions (C3D) | P3D ResNet · Sports-1M | video hit@1 (%): 61.1% -> 66.4% | Improved |
| use P3D ResNet representation instead of ResNet-152 | linear SVM features · UCF101 (3 splits, frames only) | accuracy (%): 83.5% -> 88.6% | Improved |
| fuse P3D ResNet with IDT | P3D ResNet · UCF101 (3 splits) | accuracy (%): 88.6% -> 93.7% | Improved |
| encode res5c activations instead of pool5 | P3D ResNet · UCF101 (3 splits) | accuracy (%): 88.6% -> 90.5% | Improved |
| use P3D ResNet instead of ResNet-152 features | linear SVM features · ActivityNet v1.3 validation | Top-1 accuracy (%): 71.43% -> 75.12% | Improved |
Closest papers by meaning
- Rethinking Spatiotemporal Feature Learning: Speed-Accuracy Trade-offs in Video Classification (ECCV 2018 · similarity 0.94)
- 4D Spatio-Temporal ConvNets: Minkowski Convolutional Neural Networks (CVPR 2019 · similarity 0.92)
- Can Spatiotemporal 3D CNNs Retrace the History of 2D CNNs and ImageNet? (CVPR 2018 · similarity 0.91)
- Deep Analysis of CNN-Based Spatio-Temporal Representations for Action Recognition (CVPR 2021 · similarity 0.91)
- V4D: 4D Convolutional Neural Networks for Video-level Representation Learning (ICLR 2020 · similarity 0.91)
- In Defense of Image Pre-training for Spatiotemporal Recognition (ECCV 2022 · similarity 0.90)
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.
