Research map / Detection and segmentation / 3D detection and driving
Detecting Invisible People
Tarasha Khurana, Achal Dave, Deva Ramanan
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 reporting of internal linear forecasts (ablation) | DeepSORT · MOT-17 train (FRCNN) | Occluded Top-5 F1: 28.4 -> 29.8 | Improved |
| add camera motion (egomotion) compensation (ablation) | DeepSORT + Forecast · MOT-17 train (FRCNN) | Occluded Top-5 F1: 29.8 -> 32.2 | Improved |
| add depth-based freespace filtering (ablation) | DeepSORT + Forecast + Egomotion · MOT-17 train (FRCNN) | Occluded Top-5 F1: 32.2 -> 35.7 | Improved |
| add depth-aware process noise (ablation) | DeepSORT + Forecast + Egomotion + Freespace · MOT-17 train (FRCNN) | Occluded Top-5 F1: 35.7 -> 39.8 | Improved |
| replace linear forecaster with SGAN or STGAT (ablation) | Full pipeline · MOT-17 train (FRCNN) | Occluded Top-5 F1: 39.8 -> 36.0 (SGAN); 39.8 -> 36.4 (STGAT) | Hurt |
| replace MegaDepth with Mannequin Challenge depth (ablation) | Pipeline without depth noise · MOT-17 train (FRCNN) | Occluded Top-5 F1: 35.4 -> 34.2 | Hurt |
| replace MegaDepth with MIDAS depth at default resolution (ablation) | Pipeline without depth noise · MOT-17 train (FRCNN) | Occluded Top-5 F1: 35.4 -> 34.4 | Hurt |
| increase MIDAS input resolution from 1x to 3x (ablation) | Pipeline with MIDAS depth · MOT-17 train (FRCNN) | Occluded Top-5 F1: 34.4 -> 37.5 | Improved |
| use Mask R-CNN masks instead of boxes for average depth (ablation) | Full pipeline · MOT-17 train (FRCNN) | Occluded Top-5 F1: 39.8 -> 40.6 | Improved |
| apply method on top of existing trackers | Six trackers on MOT-17 train · MOT-17 train | All Top-5 F1: -0.3 (MIFT), -0.4 (CenterTrack), +0.3 (SDP) | No clear effect |
| apply method on top of MIFT | MIFT · MOT-17 test | Occluded Top-5 F1: 38.4 -> 43.4 | Improved |
| apply method on top of Tracktor++ | Tracktor++ · MOT-20 test | Occluded Top-5 F1: 44.2 -> 46.9 | Improved |
Closest papers by meaning
- Learning To Track With Object Permanence (ICCV 2021 · similarity 0.90)
- MOTChallenge: A Benchmark for Single-Camera Multiple Target Tracking (IJCV 2020 · similarity 0.88)
- TrackFlow: Multi-Object tracking with Normalizing Flows (ICCV 2023 · similarity 0.87)
- TAO: A Large-Scale Benchmark for Tracking Any Object (ECCV 2020 · similarity 0.87)
- Believing is Seeing: Unobserved Object Detection using Generative Models (CVPR 2025 · similarity 0.86)
- CorrBEV: Multi-View 3D Object Detection by Correlation Learning with Multi-modal Prototypes (CVPR 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.
