Research map / Detection and segmentation / Weakly supervised detection and segmentation
Weakly-Supervised Cell Tracking via Backward-and-Forward Propagation
Kazuya Nishimura, Junya Hayashida, Chenyang Wang, Dai Fei Elmer Ker, Ryoma Bise
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 masked loss that ignores unassociated regions when training the tracking CNN (ablation) | Ours (MPM-Net on pseudo-labels) · Myoblast tracking (average) | association accuracy: 0.954 -> 0.960 | Improved |
| add masked loss that ignores unassociated regions when training the tracking CNN (ablation) | Ours (MPM-Net on pseudo-labels) · Myoblast tracking (average) | target effectiveness: 0.873 -> 0.881 | Improved |
| retrain MPM-Net on BF-prop pseudo-labels instead of using BF-prop output directly (ablation) | BF-prop · Myoblast tracking (BMP2 dense) | association accuracy: 0.840 -> 0.970 | Improved |
| retrain MPM-Net on BF-prop pseudo-labels instead of using BF-prop output directly (ablation) | BF-prop · Myoblast tracking (FGF2) | association accuracy: 0.794 -> 0.773 | Hurt |
| retrain MPM-Net on BF-prop pseudo-labels instead of using BF-prop output directly (ablation) | BF-prop · Myoblast tracking (Control) | target effectiveness: 0.765 -> 0.800 | Improved |
| use two-frame co-detection CNN instead of single-image detection network (ablation) | cell detection · Cell detection (FGF2+BMP2) | F1-score: 0.962 -> 0.945 | Hurt |
| use two-frame co-detection CNN instead of single-image detection network (ablation) | cell detection · Cell detection (Control) | F1-score: 0.922 -> 0.923 | No clear effect |
| increase frame interval from 5 to 25 minutes (ablation) | BF-prop association · Association (FGF2) | F1-score: 0.901 -> 0.830 | Hurt |
| increase frame interval from 5 to 25 minutes (ablation) | co-detection CNN · Cell detection (Control) | F1-score: 0.923 -> 0.926 | No clear effect |
| use weak supervision (detection points only) instead of full supervision with the same MPM-Net | MPM-Net · Myoblast tracking (FGF2) | association accuracy: 0.947 -> 0.955 | Improved |
| use weak supervision (detection points only) instead of full supervision with the same MPM-Net | MPM-Net · Myoblast tracking (Control) | association accuracy: 0.952 -> 0.926 | Hurt |
Closest papers by meaning
- Trackastra: Transformer-based cell tracking for live-cell microscopy (ECCV 2024 · similarity 0.87)
- Weakly Supervised Learning of Single-Cell Feature Embeddings (CVPR 2018 · similarity 0.86)
- Graph Neural Network for Cell Tracking in Microscopy Videos (ECCV 2022 · similarity 0.85)
- MPM: Joint Representation of Motion and Position Map for Cell Tracking (CVPR 2020 · similarity 0.84)
- Hedging Deep Features for Visual Tracking (TPAMI 2018 · similarity 0.84)
- Higher-Order Cell Tracking Transformer (NeurIPS 2026 · similarity 0.83)
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.
