Research map / Detection and segmentation / Tracking
Multi-object Tracking with Neural Gating Using Bilinear LSTM
Chanho Kim, Fuxin Li, James M. Rehg
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 |
|---|---|---|---|
| use Bilinear LSTM instead of baseline architecture 1 (ablation) | appearance gating network · MOT17 val (MOT17-09, -10), Faster R-CNN, appearance gating | IDF1: 51.28 -> 59.07 | Improved |
| use Bilinear LSTM instead of baseline architecture 2 (ablation) | appearance gating network · MOT17 val (MOT17-09, -10), Faster R-CNN, appearance gating | IDF1: 51.49 -> 59.07 | Improved |
| use Bilinear LSTM instead of baseline 1, identity switches (ablation) | appearance gating network · MOT17 val (MOT17-09, -10), Faster R-CNN, appearance gating | IDS: 412 -> 233 | Improved |
| use Bilinear LSTM instead of baseline 1, MOTA (ablation) | appearance gating network · MOT17 val (MOT17-09, -10), Faster R-CNN, appearance gating | MOTA: 50.43 -> 52.33 | Improved |
| raise LSTM state dimension from 512 to 2048 (ablation) | Bilinear LSTM · MOT17 val (MOT17-09, -10), Faster R-CNN, appearance gating | IDF1: 56.66 -> 59.07 | Improved |
| raise LSTM state dimension from 1024 to 2048 (ablation) | Bilinear LSTM · MOT17 val (MOT17-09, -10), Faster R-CNN, appearance gating | IDF1: 55.85 -> 59.07 | Improved |
| raise LSTM state dimension from 512 to 2048 (ablation) | Bilinear LSTM · MOT17 val (MOT17-09, -10), Faster R-CNN, appearance gating | MOTA: 52.14 -> 52.33 | No clear effect |
| increase max training sequence length from 10 to 40 (ablation) | Bilinear LSTM · MOT17 val (MOT17-09, -10), Faster R-CNN, appearance gating | IDF1: 54.36 -> 59.07 | Improved |
| increase max training sequence length from 40 to 160 (ablation) | Bilinear LSTM · MOT17 val (MOT17-09, -10), Faster R-CNN, appearance gating | IDF1: 59.07 -> 55.19 | Hurt |
| add re-ID pre-training to appearance gating (ablation) | Bilinear LSTM (A) · MOT17 val (MOT17-09, -10), Faster R-CNN, Nmax=40 | IDF1: 57.46 -> 59.07 | Improved |
| add re-ID pre-training to motion+appearance gating (ablation) | (M)+(A) LSTM · MOT17 val (MOT17-09, -10), Faster R-CNN, Nmax=40 | MOTA: 50.31 -> 52.63 | Improved |
| add missing detection augmentation to motion gating (ablation) | Motion LSTM · MOT17 val (MOT17-09, -10), Faster R-CNN, Nmax=40 | IDF1: 47.71 -> 50.22 | Improved |
Closest papers by meaning
- Tracking the Untrackable: Learning to Track Multiple Cues With Long-Term Dependencies (ICCV 2017 · similarity 0.88)
- Online Multi-Target Tracking Using Recurrent Neural Networks (AAAI 2017 · similarity 0.87)
- Discriminative Appearance Modeling With Multi-Track Pooling for Real-Time Multi-Object Tracking (CVPR 2021 · similarity 0.87)
- Learning Dynamic Memory Networks for Object Tracking (ECCV 2018 · similarity 0.87)
- Learning Regression and Verification Networks for Robust Long-term Tracking (IJCV 2021 · similarity 0.86)
- OVTR: End-to-End Open-Vocabulary Multiple Object Tracking with Transformer (ICLR 2025 · 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.
