Research map / Video / Action recognition
Geometry Guided Convolutional Neural Networks for Self-Supervised Video 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 |
|---|---|---|---|
| pre-train with pixel-level geometry cues from synthetic images and 3D movies | Geometry guided CNN · YUPENN | accuracy (%): 51.53 -> 69.23 | Helped |
| pre-train with pixel-level geometry cues from synthetic images and 3D movies | Geometry guided CNN · Maryland | accuracy (%): 76.67 -> 86.90 | Helped |
| pre-train with geometry cues | video action recognition CNN · UCF101 | accuracy (%): 50.9 -> 54.1 | Helped |
| pre-train with geometry cues | video action recognition CNN · HMDB51 | accuracy (%): 19.8 -> 22.6 | Helped |
| add 3D movie geometry pre-training after FlyingChairs pre-training (ablation) | Geometry guided CNN · UCF101 | accuracy (%): 50.2 -> 54.1 | Helped |
| add 3D movie geometry pre-training after FlyingChairs pre-training (ablation) | Geometry guided CNN · HMDB51 | accuracy (%): 19.1 -> 22.6 | Helped |
| use progressive training rather than pooling both geometry datasets together (ablation) | Geometry guided CNN · UCF101 | accuracy (%): 52.4 -> 54.1 | Helped |
| use progressive training rather than averaging separately trained models (ablation) | Geometry guided CNN · UCF101 | accuracy (%): 52.6 -> 54.1 | Helped |
| retain the learning-without-forgetting term when training on 3D movies (ablation) | Geometry guided CNN · UCF101 | accuracy (%): 50.0 -> 54.1 | Helped |
| train on accurate FlyingChairs geometry before the 3D movies (ablation) | Geometry guided CNN · UCF101 | accuracy (%): 52.9 -> 54.1 | Helped |
| add geometry-model scores to ImageNet model scores | ImageNet-pretrained CNN · UCF101 | accuracy (%): 63.3 -> 66.1 | Helped |
| add geometry-model scores to ImageNet model scores | ImageNet-pretrained CNN · HMDB51 | accuracy (%): 28.5 -> 30.7 | Helped |
Closest papers by meaning
- Learning Spatial Common Sense With Geometry-Aware Recurrent Networks (CVPR 2019 · similarity 0.88)
- Semantically-Guided Representation Learning for Self-Supervised Monocular Depth (ICLR 2020 · similarity 0.88)
- Self-Supervised Neural Articulated Shape and Appearance Models (CVPR 2022 · similarity 0.88)
- Self-Supervised Spatio-Temporal Representation Learning for Videos by Predicting Motion and Appearance Statistics (CVPR 2019 · similarity 0.87)
- Revisiting Self-Supervised Visual Representation Learning (CVPR 2019 · similarity 0.87)
- Self-Supervised Learning of Interpretable Keypoints From Unlabelled Videos (CVPR 2020 · similarity 0.87)
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.