Research map / Detection and segmentation / CNN features
Highly Efficient and Unsupervised Framework for Moving Object Detection in Satellite Videos
Chao Xiao, Wei An, Yifan Zhang, Zhuo Su, Miao Li, Weidong Sheng, Matti Pietikäinen, Li Liu
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 |
|---|---|---|---|
| train sparse network on manual labels instead of DSFNet | HiEUM-sup vs DSFNet · Jilin-1 satellite videos (relabeled test set) | average F1: 74.4 -> 82.1 | Improved |
| replace manual labels with traditional-method labels (ablation) | HiEUM · Jilin-1 satellite videos (relabeled test set) | average F1: 82.1 -> 74.1 | Hurt |
| add label self-evolution with trajectory filtering (ablation) | HiEUM-unsup · Jilin-1 satellite videos (relabeled test set) | average F1: 74.1 -> 89.7 | Improved |
| add sparse network on top of adaptive-threshold frame differencing (ablation) | HiEUM · Jilin-1 satellite videos (relabeled test set) | average F1: 66.1 -> 89.7 | Improved |
| add absolute value of residuals before threshold (ablation) | HiEUM · Jilin-1 satellite videos (relabeled test set) | average F1: 75.0 -> 89.7 | Improved |
| raise threshold k from 3 to 15 (ablation) | HiEUM · Jilin-1 satellite videos (relabeled test set) | average F1: 89.7 -> 65.3 | Hurt |
| lower threshold k from 3 to 1 (ablation) | HiEUM · Jilin-1 satellite videos (relabeled test set) | average F1: 89.7 -> 82.3 | Hurt |
| raise threshold k from 1 to 15 (speed) (ablation) | HiEUM · Jilin-1 satellite videos (relabeled test set) | FPS: 20.8 -> 248.4 | Improved |
| increase input frames from 10 to 20 (ablation) | HiEUM · Jilin-1 satellite videos (relabeled test set) | average F1: 82.6 -> 89.7 | Improved |
| increase input frames from 20 to 60 (ablation) | HiEUM · Jilin-1 satellite videos (relabeled test set) | average F1: 89.7 -> 88.8 | Hurt |
| increase sparse U-net depth from 2 to 3 layers (ablation) | HiEUM · Jilin-1 satellite videos (relabeled test set) | average F1: 89.2 -> 89.7 | Improved |
| increase sparse U-net depth from 3 to 4 layers (ablation) | HiEUM · Jilin-1 satellite videos (relabeled test set) | average F1: 89.7 -> 89.6 | No clear effect |
Closest papers by meaning
- Moving Vehicle Detection for Remote Sensing Video Surveillance with Nonstationary Satellite Platform (TPAMI 2021 · similarity 0.86)
- Moving Object Detection in Time-Lapse or Motion Trigger Image Sequences Using Low-Rank and Invariant Sparse Decomposition (ICCV 2017 · similarity 0.85)
- S²Teacher: Step-by-step Teacher for Sparsely Annotated Oriented Object Detection (AAAI 2026 · similarity 0.85)
- VISO: Accelerating In-orbit Object Detection with Language-Guided Mask Learning and Sparse Inference (ICCV 2025 · similarity 0.85)
- LEOD: Label-Efficient Object Detection for Event Cameras (CVPR 2024 · similarity 0.84)
- SimLTD: Simple Supervised and Semi-Supervised Long-Tailed Object Detection (CVPR 2025 · 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.
