Research map / Detection and segmentation / CNN features
Dynamic Zoom-In Network for Fast Object Detection in Large Images
Read the paperCheck 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 |
|---|---|---|---|
| Use sequential Q-net selection instead of greedy selection | Dynamic Zoom-In Network · CPD | Aperc: 65% -> 99% | Helped |
| Estimate zoom utility with R-net rather than coarse-detector entropy | Qnet*-CNN · CPD | Aperc: 88% -> 99% | Helped |
| Add window refinement to adjust selected zoom locations (ablation) | Qnet-CNN+Rnet · WP (Pperc ≤30%) | Aperc: 83% -> 92% | Helped |
| Add window refinement to adjust selected zoom locations (ablation) | Qnet-CNN+Rnet · CPD (Pperc ≤40%) | Aperc: 65% -> 99% | Helped |
| Use a convolutional Q-net instead of a fully connected Q-net (ablation) | Qnet*-CNN+Rnet · WP (Pperc ≤30%) | Aperc: 71% -> 92% | Helped |
| Use a fully connected Q-net to exploit driving-scene spatial priors (ablation) | Qnet*-CNN+Rnet · CPD (Pperc ≤40%) | Aperc: 99% -> 93% | Hurt |
| Include the explicit cost penalty in the reward (ablation) | Qnet*-CNN+Rnet* · CPD (Pperc ≤40%) | Aperc: 65% -> 99% | Helped |
| Remove explicit cost penalty from the reward (ablation) | Qnet*-CNN+Rnet · WP (Pperc ≤30%) | Aperc: 92% -> 93% | No clear effect |
| Downsample without any zoom-in processing instead of using adaptive zooms | Fine-detection-all · CPD | AP: 0.493 -> 0.322 | Hurt |
| Downsample without any zoom-in processing instead of using adaptive zooms | Fine-detection-all · WP | AP: 0.407 -> 0.289 | Hurt |
| Apply dynamic zooming rather than an off-the-shelf single-shot detector | Object detection on large images · WP | AP: 0.255 -> 0.379 | Helped |
| Use dynamic zooming rather than an off-the-shelf single-shot detector | Object detection on large images · CPD | AP: 0.398 -> 0.503 | Helped |
Closest papers by meaning
- Optimizing Video Object Detection via a Scale-Time Lattice (CVPR 2018 · similarity 0.88)
- Reinforcement Learning for Visual Object Detection (CVPR 2016 · similarity 0.88)
- Deep Reinforcement Learning of Region Proposal Networks for Object Detection (CVPR 2018 · similarity 0.88)
- Adaptive Object Detection Using Adjacency and Zoom Prediction (CVPR 2016 · similarity 0.88)
- R-FCN: Object Detection via Region-based Fully Convolutional Networks (NeurIPS 2016 · similarity 0.87)
- Training-Time-Friendly Network for Real-Time Object Detection (AAAI 2020 · 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.