Research map / Detection and segmentation / CNN features
Context R-CNN: Long Term Temporal Context for Per-Camera Object Detection
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 long- and short-term per-camera attention context | Faster R-CNN ResNet-101 · Snapshot Serengeti (SS) | mAP: 37.9 -> 55.9 | Helped |
| add long- and short-term per-camera attention context | Faster R-CNN ResNet-101 · Caltech Camera Traps (CCT) | mAP: 56.8 -> 76.3 | Helped |
| add long- and short-term per-camera attention context | Faster R-CNN ResNet-101 · CityCam (CC) | mAP: 38.1 -> 42.6 | Helped |
| replace frame-level detection with majority voting across the 3-frame window | Faster R-CNN · Snapshot Serengeti (SS) | mAP: 37.9 -> 37.8 | No clear effect |
| add spatially aligned short-term feature averaging | Faster R-CNN · Snapshot Serengeti (SS) | mAP: 37.9 -> 39.6 | Helped |
| add non-local attention among proposals in the current frame (ablation) | Faster R-CNN · Snapshot Serengeti (SS) | mAP: 37.9 -> 44.9 | Helped |
| add nearby-frame short-term attention to current-frame attention (ablation) | SF Attn · Snapshot Serengeti (SS) | mAP: 44.9 -> 46.4 | Helped |
| add month-horizon long-term attention to short-term attention (ablation) | ST Attn · Snapshot Serengeti (SS) | mAP: 46.4 -> 55.6 | Helped |
| extend long-term context horizon from one minute to one month (ablation) | LT Attn · Snapshot Serengeti (SS) | mAP: 50.3 -> 55.6 | Helped |
| use dataset-fine-tuned rather than COCO-only memory features (ablation) | Context R-CNN · Snapshot Serengeti (SS) | mAP: 50.3 -> 55.6 | Helped |
| retain empty-image representations in long-term memory (ablation) | Context R-CNN · Snapshot Serengeti (SS) | mAP: 53.9 -> 55.6 | Helped |
| reduce stored long-term memory representations by half (ablation) | Context R-CNN · Snapshot Serengeti (SS) | mAP: 55.6 -> 52.5 | Hurt |
Closest papers by meaning
- ContextShift: A Controlled Benchmark for Context Dependence in Object Detection (NeurIPS 2026 · similarity 0.88)
- Seeing without Looking: Contextual Rescoring of Object Detections for AP Maximization (CVPR 2020 · similarity 0.88)
- Contextual Convolutional Networks (ICLR 2023 · similarity 0.88)
- Context Enhanced Transformer for Single Image Object Detection in Video Data (AAAI 2024 · similarity 0.88)
- Rethinking Scale-Aware Temporal Encoding for Event-based Object Detection (NeurIPS 2025 · similarity 0.88)
- Mobile Video Object Detection With Temporally-Aware Feature Maps (CVPR 2018 · similarity 0.87)
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.