Research map / Detection and segmentation / CNN features
Progressive Sparse Local Attention for Video Object Detection
Chaoxu Guo, Bin Fan, Jie Gu, Qian Zhang, Shiming Xiang, Veronique Prinet, Chunhong Pan
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 Recursive Feature Updating (RFU) (ablation) | PSLA (a) · ImageNet VID val | mAP: 74.4 -> 75.8 | Improved |
| add Dense Feature Transforming Quality Net (ablation) | PSLA (b) · ImageNet VID val | mAP: 75.8 -> 77.1 | Improved |
| add RFU and Quality Net to Nonlocal propagation (ablation) | Nonlocal_S · ImageNet VID val | mAP: 72.1 -> 74.1 | Improved |
| add RFU and Quality Net to MatchTrans propagation (ablation) | MatchTrans_S · ImageNet VID val | mAP: 72.5 -> 75.2 | Improved |
| add RFU and Quality Net to DensePSLA propagation (ablation) | DensePSLA_S · ImageNet VID val | mAP: 73.6 -> 75.7 | Improved |
| use progressively sparse local region instead of dense local region (ablation) | DensePSLA_S · ImageNet VID val (max displacement 4) | mAP: 73.6 -> 74.4 | Improved |
| use local progressive sparse attention instead of global Nonlocal (ablation) | Nonlocal_S · ImageNet VID val | mAP: 72.1 -> 74.4 | Improved |
| use softmax-normalized dense local PSLA instead of MatchTrans accumulation (ablation) | MatchTrans_S · ImageNet VID val (max displacement 3) | mAP: 71.5 -> 73.7 | Improved |
| replace optical flow propagation (DFF) with PSLA, key frame interval 35 | DFF re-implementation · ImageNet VID val | mAP: 67.5 -> 68.5 | Improved |
| replace optical flow propagation (DFF) with PSLA, key frame interval 15 | DFF re-implementation · ImageNet VID val | mAP: 72.2 -> 72.9 | Improved |
| increase max displacement from 2 to 4 | PSLA (a) · ImageNet VID val | mAP: 73.6 -> 74.4 | Improved |
| add DCN to the ResNet101 backbone | PSLA · ImageNet VID val | mAP: 77.1 -> 80.0 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- R-FCN — R-FCN: Object Detection via Region-based Fully Convolutional Networks (NeurIPS 2016) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Low-latency Event-based Object Detection with Spatially-Sparse Linear Attention (ECCV 2026 · similarity 0.87)
- Learning Where to Focus for Efficient Video Object Detection (ECCV 2020 · similarity 0.86)
- Sparse R-CNN: An End-to-End Framework for Object Detection (TPAMI 2023 · similarity 0.86)
- SMamba: Sparse Mamba for Event-based Object Detection (AAAI 2025 · similarity 0.85)
- SparseFormer: Sparse Visual Recognition via Limited Latent Tokens (ICLR 2024 · similarity 0.84)
- Scene Adaptive Sparse Transformer for Event-based Object Detection (CVPR 2024 · similarity 0.84)
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.
