# Dynamic Zoom-in Network for Fast Object Detection in Large Images

arXiv: https://arxiv.org/abs/1711.05187 · HTML: https://mlsanity.com/paper/1711.05187/

## Key facts

- Cited by 23 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 14 Nov 2017.

## Built on by (most cited first, 23 of 23)

- [AdaMML: Adaptive Multi-Modal Learning for Efficient Video Recognition](https://mlsanity.com/paper/2105.05165/) (2021)
- [FOVEA: Foveated Image Magnification for Autonomous Navigation](https://mlsanity.com/paper/2108.12102/) (2021)
- [DeepPerform: An Efficient Approach for Performance Testing of Resource-Constrained Neural Networks](https://mlsanity.com/paper/2210.05370/) (2022)
- [OmniCount: Multi-label Object Counting with Semantic-Geometric Priors](https://mlsanity.com/paper/2403.05435/) (2024)
- [SALISA: Saliency-based Input Sampling for Efficient Video Object Detection](https://mlsanity.com/paper/2204.02397/) (2022)
- [AdaZoom: Adaptive Zoom Network for Multi-Scale Object Detection in Large Scenes](https://mlsanity.com/paper/2106.10409/) (2021)
- [FAQ: Feature Aggregated Queries for Transformer-based Video Object Detectors](https://mlsanity.com/paper/2303.08319/) (2023)
- [Efficient High-Resolution Deep Learning: A Survey](https://mlsanity.com/paper/2207.13050/) (2022)
- [DyCL: Dynamic Neural Network Compilation Via Program Rewriting and Graph Optimization](https://mlsanity.com/paper/2307.04963/) (2023)
- [Survey: Exploiting Data Redundancy for Optimization of Deep Learning](https://mlsanity.com/paper/2208.13363/) (2022)
- [Fewer is More: Efficient Object Detection in Large Aerial Images](https://mlsanity.com/paper/2212.13136/) (2022)
- [Improved Techniques for Quantizing Deep Networks with Adaptive Bit-Widths](https://mlsanity.com/paper/2103.01435/) (2021)
- [Universal Object Detection with Large Vision Model](https://mlsanity.com/paper/2212.09408/) (2022)
- [SaccadeDet: A Novel Dual-Stage Architecture for Rapid and Accurate Detection in Gigapixel Images](https://mlsanity.com/paper/2407.17956/) (2024)
- [Aerial View Localization with Reinforcement Learning: Towards Emulating Search-and-Rescue](https://mlsanity.com/paper/2209.03694/) (2022)
- [3D Small Object Detection with Dynamic Spatial Pruning](https://mlsanity.com/paper/2305.03716/) (2023)
- [EgoAdapt: Adaptive Multisensory Distillation and Policy Learning for Efficient Egocentric Perception](https://mlsanity.com/paper/2506.21080/) (2025)
- [You Better Look Twice: a new perspective for designing accurate detectors with reduced computations](https://mlsanity.com/paper/2107.10050/) (2021)
- [Problem-dependent attention and effort in neural networks with applications to image resolution and model selection](https://mlsanity.com/paper/2201.01415/) (2022)
- [A Coarse to Fine Framework for Object Detection in High Resolution Image](https://mlsanity.com/paper/2303.01219/) (2023)
- [The Untapped Potential of Off-the-Shelf Convolutional Neural Networks](https://mlsanity.com/paper/2103.09891/) (2021)
- [Identifying Light-curve Signals with a Deep Learning Based Object Detection Algorithm. II. A General Light Curve Classification Framework](https://mlsanity.com/paper/2311.08080/) (2023)
- [HiResNets: Native Full-HD Video Recognition with Foveal Residual Streams](https://mlsanity.com/paper/2608.02140/) (2026)

## Abstract (arXiv)

We introduce a generic framework that reduces the computational cost of object detection while retaining accuracy for scenarios where objects with varied sizes appear in high resolution images. Detection progresses in a coarse-to-fine manner, first on a down-sampled version of the image and then on a sequence of higher resolution regions identified as likely to improve the detection accuracy. Built upon reinforcement learning, our approach consists of a model (R-net) that uses coarse detection results to predict the potential accuracy gain for analyzing a region at a higher resolution and another model (Q-net) that sequentially selects regions to zoom in. Experiments on the Caltech Pedestrians dataset show that our approach reduces the number of processed pixels by over 50% without a drop in detection accuracy. The merits of our approach become more significant on a high resolution test set collected from YFCC100M dataset, where our approach maintains high detection performance while reducing the number of processed pixels by about 70% and the detection time by over 50%.