Abstract:Detecting oriented tiny objects, which are limited in appearance information yet prevalent in real-world applications, remains an intricate and under-explored problem. To address this, we systemically introduce a new dataset, benchmark, and a dynamic coarse-to-fine learning scheme in this study. Our proposed dataset, AI-TOD-R, features the smallest object sizes among all oriented object detection datasets. Based on AI-TOD-R, we present a benchmark spanning a broad range of detection paradigms, including both fully-supervised and label-efficient approaches. Through investigation, we identify a learning bias presents across various learning pipelines: confident objects become increasingly confident, while vulnerable oriented tiny objects are further marginalized, hindering their detection performance. To mitigate this issue, we propose a Dynamic Coarse-to-Fine Learning (DCFL) scheme to achieve unbiased learning. DCFL dynamically updates prior positions to better align with the limited areas of oriented tiny objects, and it assigns samples in a way that balances both quantity and quality across different object shapes, thus mitigating biases in prior settings and sample selection. Extensive experiments across eight challenging object detection datasets demonstrate that DCFL achieves state-of-the-art accuracy, high efficiency, and remarkable versatility. The dataset, benchmark, and code are available at <a class="link-external link-https" href="https://chasel-tsui.github.io/AI-TOD-R/" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the highly challenging problem of **oriented tiny object detection**. Specifically, the paper mainly focuses on the following points: 1. **Lack of datasets and benchmarks**: Existing research and datasets mainly focus on general object detection or object detection in arbitrary directions, but for the task of detecting extremely small targets with orientation information, there are still a lack of specialized datasets and benchmarks. This has restricted the development of related fields. 2. **Learning bias**: When dealing with extremely small targets, existing methods usually show learning bias. Specifically, the model is more confident in high - confidence targets, while further marginalizing extremely small targets that are difficult to detect, thus affecting the detection performance. This bias is common in different learning pipelines, especially during the training process. Due to the limited feature information of extremely small targets, they are easily suppressed or ignored. 3. **Improving detection performance**: To address the above challenges, the paper proposes a new Dynamic Coarse - to - Fine Learning (DCFL) scheme, aiming to achieve unbiased learning by dynamically updating prior positions and balancing the quantity and quality of samples at different scales, thereby improving the detection performance of extremely small targets. ### Main contributions of the paper - **New dataset AI - TOD - R**: A new dataset AI - TOD - R designed specifically for the detection of small targets in arbitrary directions is introduced. Its average target size is only 10.62 pixels, making it one of the datasets with the smallest target sizes currently. - **Comprehensive benchmarking**: Based on the AI - TOD - R dataset, a benchmark covering multiple detection paradigms is constructed, including fully - supervised and label - efficient detection methods, revealing the learning biases of different methods when dealing with extremely small targets. - **DCFL method**: A Dynamic Coarse - to - Fine Learning (DCFL) scheme is proposed. Through adaptive updating of prior positions and a phased sample learning strategy, the learning bias problem in extremely small target detection is effectively alleviated, and the detection performance is significantly improved. ### Conclusion Through these contributions, the paper not only fills the resource gap in the field of extremely small target detection but also provides an effective solution to address the key challenges in this field, promoting the progress of related research.

Oriented Tiny Object Detection: A Dataset, Benchmark, and Dynamic Unbiased Learning

Dynamic Coarse-to-Fine Learning for Oriented Tiny Object Detection

Detecting tiny objects in aerial images: A normalized Wasserstein distance and a new benchmark

Dense Tiny Object Detection: A Scene Context Guided Approach and a Unified Benchmark

Oriented RepPoints for Aerial Object Detection

Adaptive Dynamic Label Assignment for Tiny Object Detection in Aerial Images

A Survey of the Four Pillars for Small Object Detection: Multiscale Representation, Contextual Information, Super-Resolution, and Region Proposal

Highly Efficient Anchor-Free Oriented Small Object Detection for Remote Sensing Images via Periodic Pseudo-Domain

Oriented Object Detection via Contextual Dependence Mining and Penalty-Incentive Allocation

The 1st Tiny Object Detection Challenge:Methods and Results

Align and Distill: Unifying and Improving Domain Adaptive Object Detection

Described Object Detection: Liberating Object Detection with Flexible Expressions

Oriented R-CNN and Beyond

Tiny-DSOD: Lightweight Object Detection for Resource-Restricted Usages

Robust Tiny Object Detection in Aerial Images amidst Label Noise

Anchor-Free Oriented Proposal Generator for Object Detection

Towards RAW Object Detection in Diverse Conditions

Oriented R-CNN for Object Detection

Oriented Object Detection in Optical Remote Sensing Images using Deep Learning: A Survey

Regression-Guided Refocusing Learning With Feature Alignment for Remote Sensing Tiny Object Detection

Visible-Thermal Tiny Object Detection: A Benchmark Dataset and Baselines