Research map / Video / Action recognition
Unsupervised Learning of Action Classes With Continuous Temporal Embedding
Check 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 |
|---|---|---|---|
| Use the continuous temporal embedding instead of ImageNet, I3D, dense trajectory, video vector, or video Darwin representations (ablation) | Unsupervised temporal action segmentation · Breakfast | MoF: 21.2% -> 41.8%; 25.1% -> 41.8%; 31.6% -> 41.8%; 30.1% -> 41.8%; 36.6% -> 41.8% | Helped |
| Use Viterbi decoding instead of Mallow plus multinomial decoding (ablation) | Continuous temporal embedding · Breakfast | MoF: 29.5% -> 41.8% | Helped |
| Add Viterbi decoding rather than using Mallow decoding alone (ablation) | Continuous temporal embedding · Breakfast | MoF: 29.5% -> 34.8% | Helped |
| Add Mallow reordering to Viterbi decoding (ablation) | Continuous temporal embedding · Breakfast | MoF: 41.8% -> 34.8% | Hurt |
| Pair Rankloss embedding with Viterbi instead of Mallow decoding (ablation) | Rankloss embedding · Breakfast | MoF: 34.6% -> 27.1% | Hurt |
| Add bag-of-words quantization to video clustering (ablation) | Unknown-activity pipeline · Breakfast | Mean over videos: 19.3% -> 31.8% | Helped |
| Use soft instead of hard bag-of-words assignment (ablation) | Unknown-activity pipeline · Breakfast | Mean over videos: 29.8% -> 31.8% | Helped |
| Use one embedding for the full dataset instead of learning additional embeddings for each video cluster (ablation) | Unknown-activity pipeline · Breakfast | MoF: 16.4% -> 18.3% | Helped |
| Add activity-level clustering rather than cluster all videos as one activity group (ablation) | Unknown-activity pipeline · Breakfast | MoF: 10.7% -> 18.3% | Helped |
| Use activity clustering and temporal embedding rather than direct k-Means on video features (ablation) | Unknown-activity pipeline · Breakfast | MoF: 6.1% -> 18.3% | Helped |
| Increase the background ratio τ from 60 to 99 (ablation) | Background model · YouTube Instructions | IoU with background: 4.9% -> 2.7% | Hurt |
| Increase the background ratio τ from 75 to 99 (ablation) | Background model · YouTube Instructions | MoF with background: 10.1% -> 58.6% | Helped |
Closest papers by meaning
- Action Shuffle Alternating Learning for Unsupervised Action Segmentation (CVPR 2021 · similarity 0.91)
- Unsupervised Action Segmentation via Fast Learning of Semantically Consistent Actoms (AAAI 2024 · similarity 0.90)
- Hierarchical Vector Quantization for Unsupervised Action Segmentation (AAAI 2025 · similarity 0.90)
- Unsupervised Action Segmentation by Joint Representation Learning and Online Clustering (CVPR 2022 · similarity 0.89)
- Unsupervised Deep Learning of Mid-Level Video Representation for Action Recognition (AAAI 2018 · similarity 0.89)
- Temporally Consistent Unbalanced Optimal Transport for Unsupervised Action Segmentation (CVPR 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.