Research map / Video / Video segmentation and tracking
Exploiting Temporal Coherence for Self-Supervised One-shot Video Re-identification
Dripta S. Raychaudhuri, Amit K. Roy-Chowdhury
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 |
|---|---|---|---|
| add the inter-sequence loss to intra-sequence-only TCPL (ablation) | TCPL · MARS (one-shot, p=0.05) | Rank-1: 63.3 -> 65.2 | Improved |
| add the intra-sequence loss to inter-sequence-only TCPL (ablation) | TCPL · DukeMTMC-VideoReID (one-shot, p=0.05) | Rank-1: 74.4 -> 76.8 | Improved |
| add the inter-sequence loss to intra-sequence-only TCPL (ablation) | TCPL · DukeMTMC-VideoReID (one-shot, p=0.05) | mAP: 67.7 -> 67.9 | Improved |
| add temporal consistency losses to progressive pseudo-labeling (TCPL vs One-Shot Prog.) | One-Shot Prog. · DukeMTMC-VideoReID (one-shot, p=0.05) | Rank-1: 72.9 -> 76.8 | Improved |
| add temporal consistency losses to progressive pseudo-labeling (TCPL vs One-Shot Prog.) | One-Shot Prog. · MARS (one-shot, p=0.05) | Rank-1: 62.8 -> 65.2 | Improved |
| use aggressive enlarging factor p=0.20 instead of 0.05 with intra-sequence loss only (ablation) | TCPL with L_intra only · DukeMTMC-VideoReID (one-shot) | Rank-1: 76.2 -> 74.4 | Hurt |
| use aggressive enlarging factor p=0.20 instead of 0.05 for EUG baseline (ablation) | EUG · DukeMTMC-VideoReID (one-shot) | Rank-1: 72.8 -> 68.9 | Hurt |
| add temporal consistency losses at p=0.20 (TCPL L_intra vs One-Shot Prog.) | One-Shot Prog. · DukeMTMC-VideoReID (one-shot, p=0.20) | Rank-1: 69.1 -> 74.4 | Improved |
| raise inter-loss negative rank range r from 2 to 3 (ablation) | TCPL with L_inter only · MARS (p=0.20) | Rank-1: 53.6 -> 50.8 | Hurt |
| lower inter-loss negative rank range r from 2 to 1 (ablation) | TCPL with L_inter only · MARS (p=0.20) | Rank-1: 53.6 -> 53.1 | Hurt |
| increase labeled data from 10% to 20% (few-example) | TCPL · MARS (few-example) | Rank-1: 72.0 -> 78.2 | Improved |
| change random selection of the initial labeled tracklets (ablation) | TCPL · DukeMTMC-VideoReID (p=0.20) | Rank-1: 74.7 -> 74.4 | No clear effect |
Closest papers by meaning
- Exploit the Unknown Gradually: One-Shot Video-Based Person Re-Identification by Stepwise Learning (CVPR 2018 · similarity 0.88)
- Slow and Steady Feature Analysis: Higher Order Temporal Coherence in Video (CVPR 2016 · similarity 0.86)
- Temporal Coherence or Temporal Motion: Which is More Critical for Video-based Person Re-identification? (ECCV 2020 · similarity 0.86)
- Scene Consistency Representation Learning for Video Scene Segmentation (CVPR 2022 · similarity 0.86)
- Temporal Knowledge Propagation for Image-to-Video Person Re-Identification (ICCV 2019 · similarity 0.86)
- Unsupervised Person Re-Identification by Camera-Aware Similarity Consistency Learning (ICCV 2019 · 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.
