Research map / Video / Action recognition
Unsupervised Deep Learning of Mid-Level Video Representation for Action Recognition
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 k-means clustering of hand-crafted IT features with discriminative (SVM) clustering (ablation) | mid-level video representation · HMDB51 | accuracy (%): 54.7 -> 56.5 | Helped |
| learn features with a deep network from clusters (softmax) instead of hand-crafted features with discriminative clustering (ablation) | mid-level video representation · HMDB51 | accuracy (%): 56.5 -> 62.6 | Helped |
| use hinge+Fisher joint loss with Euclidean-compatible unified metric instead of softmax loss (ablation) | deep learning from clusters · HMDB51 | accuracy (%): 62.6 -> 66.8 | Helped |
| use hinge+Fisher joint loss with Euclidean-compatible unified metric instead of softmax loss (ablation) | deep learning from clusters · UCF101 | accuracy (%): 88.1 -> 90.3 | Helped |
| fuse features of fully connected and fully convolutional autoencoders (GMKL) | Ours (φe) · HMDB51 | accuracy (%): 66.8 -> 73.8 | Helped |
| fuse features of fully connected and fully convolutional autoencoders (GMKL) | Ours (φe) · UCF101 | accuracy (%): 90.3 -> 93.6 | Helped |
| add RGB features from ResNet-152 | Ours (φe + ϕe) · HMDB51 | accuracy (%): 73.8 -> 74.1 | Helped |
| add RGB features from ResNet-152 | Ours (φe + ϕe) · UCF101 | accuracy (%): 93.6 -> 95.3 | Helped |
| train encoder on HMDB51 instead of UCF101 itself, then extract UCF101 features (ablation) | Ours (φe) · UCF101 | accuracy (%): 90.3 -> 88.3 | Hurt |
Closest papers by meaning
- Unsupervised Learning of Action Classes With Continuous Temporal Embedding (CVPR 2019 · similarity 0.89)
- Temporally-Weighted Hierarchical Clustering for Unsupervised Action Segmentation (CVPR 2021 · similarity 0.89)
- Unsupervised Learning of View-invariant Action Representations (NeurIPS 2018 · similarity 0.89)
- No Frame Left Behind: Full Video Action Recognition (CVPR 2021 · similarity 0.89)
- Iterative Contrast-Classify for Semi-supervised Temporal Action Segmentation (AAAI 2022 · similarity 0.89)
- Unsupervised Action Segmentation via Fast Learning of Semantically Consistent Actoms (AAAI 2024 · similarity 0.88)
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.