Research map / Detection and segmentation / Tracking
STORM: Segment, Track, and Object Re-Localization from a Single Image
Yu Deng, Teng Cao, Hikaru Shindo, Quentin Delfosse, Jiahong Xue, Kristian Kersting
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 attention (1 layer) to TOM matching (ablation) | TOM · Tracking-loss verification (synthetic BOP) | accuracy (%): 95.23 -> 98.36 | Improved |
| use learned TOM without attention instead of cosine similarity (ablation) | TOM · Tracking-loss verification (synthetic BOP) | accuracy (%): 87.55 -> 95.23 | Improved |
| stack a second attention layer in TOM (ablation) | TOM · Tracking-loss verification (synthetic BOP) | accuracy (%): 98.36 -> 97.84 | No clear effect |
| add TOM tracking-loss detection to FoundationPose tracking (ablation) | STORM · Modified YCB-V tracking stress test | ADD AUC: 52.76 -> 74.64 | Improved |
| add TOM tracking-loss detection to FoundationPose tracking (ablation) | STORM · Modified YCB-V tracking stress test | AR: 50.09 -> 67.85 | Improved |
| add language conditioning to SOM (k=1) (ablation) | SOM · BOP segmentation (k=1 reference views) | AP_C: 67.3 -> 67.7 | Improved |
| add language conditioning to SOM (k=16) (ablation) | SOM · BOP segmentation (k=16 reference views) | AP_C: 67.9 -> 68.0 | Improved |
| increase reference views from 1 to 16 (with language) (ablation) | SOM · BOP segmentation | AP_C: 67.7 -> 68.0 | Improved |
| increase reference views from 1 to 16 (with language) (ablation) | SOM · BOP segmentation | latency (ms): 45.8 -> 61.8 | Hurt |
| use 7 HSFA layers instead of 3 (ablation) | SOM · SOM HSFA ablation test set | AP (%): 1 percentage point higher AP (95.4%) | No clear effect |
| use HSFA3 or HSFA7 instead of HSFA0 (simple attention, conv, linear) (ablation) | SOM · SOM HSFA ablation test set | AP (%): more than 50 percentage points higher than HSFA0 | Improved |
| align SAM3D reconstruction to benchmark coordinate frame (ablation) | SAM3D + FoundationPose · YCB-V pose with SAM3D models | VSD: 22.19 -> 60.72 | Improved |
Closest papers by meaning
- FoundationPose: Unified 6D Pose Estimation and Tracking of Novel Objects (CVPR 2024 · similarity 0.84)
- 6D Object Pose Tracking in Internet Videos for Robotic Manipulation (ICLR 2025 · similarity 0.84)
- ProxyPose: 6-DoF Pose Tracking via Video-to-Video Translation (NeurIPS 2026 · similarity 0.83)
- STORM: Spatio-TempOral Reconstruction Model For Large-Scale Outdoor Scenes (ICLR 2025 · similarity 0.83)
- SingRef6D: Monocular Novel Object Pose Estimation with a Single RGB Reference (NeurIPS 2025 · similarity 0.82)
- MIMTrack: In-Context Tracking via Masked Image Modeling (AAAI 2025 · similarity 0.82)
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.
