Research map / Video / Action recognition
NeuralNetwork-Viterbi: A Framework for Weakly Supervised Video 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 |
|---|---|---|---|
| Use direct Viterbi-based learning with an explicit length model and language model instead of pseudo ground-truth training with a language model | NN-Viterbi on Breakfast · Breakfast | accuracy (%): 39.1 -> 43.0 | Helped |
| Add an HMM to pseudo ground-truth training without a language model | Pseudo ground-truth training on Breakfast · Breakfast | accuracy (%): 23.9 -> 33.3 | Helped |
| Replace the HMM with an explicit length model and add a language model in direct learning | NN-Viterbi on Breakfast · Breakfast | accuracy (%): 36.4 -> 43.0 | Helped |
| Add buffered frames sampled from previously processed sequences (ablation) | Online NN-Viterbi · Breakfast | frame accuracy (%): 27.2% -> a huge improvement | Helped |
| Increase the buffered-frame sampling ratio from 1:1 toward 1:25 (ablation) | NN-Viterbi with buffered sampling · Breakfast | frame accuracy | Helped |
| Use fewer than 100 buffered sequences (ablation) | NN-Viterbi with buffered sampling · Breakfast | segmentation performance | Hurt |
| Increase the buffer to more than 200 sequences (ablation) | NN-Viterbi with buffered sampling · Breakfast | segmentation performance | Helped |
| Increase batch size above one (ablation) | NN-Viterbi · Breakfast | accuracy | Hurt |
| Sort training videos by coarse activity instead of randomly shuffling them (ablation) | NN-Viterbi · Breakfast | frame accuracy (%): 43.0 -> 27.9 | Hurt |
| Sort training videos by actor instead of randomly shuffling them (ablation) | NN-Viterbi · Breakfast | frame accuracy (%): 43.0 -> 41.5 | Hurt |
| Include explicit context and length modeling rather than relying on the recurrent network alone | Weakly supervised video segmentation · Breakfast and 50 Salads | frame accuracy | Helped |
Closest papers by meaning
- Set-Constrained Viterbi for Set-Supervised Action Segmentation (CVPR 2020 · similarity 0.90)
- Anchor-Constrained Viterbi for Set-Supervised Action Segmentation (CVPR 2021 · similarity 0.88)
- Weakly Supervised Instance Segmentation for Videos With Temporal Mask Consistency (CVPR 2021 · similarity 0.87)
- Beyond Short Clips: End-to-End Video-Level Learning With Collaborative Memories (CVPR 2021 · similarity 0.86)
- SCT: Set Constrained Temporal Transformer for Set Supervised Action Segmentation (CVPR 2020 · similarity 0.86)
- Weakly-Supervised Action Segmentation With Iterative Soft Boundary Assignment (CVPR 2018 · similarity 0.86)
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.