Augmentation for small object detection

By Mate Kisantal, Zbigniew Wojna, Jakub Murawski, Jacek Naruniec, Kyunghyun Cho. International Conference on Advances in Computing and Information Technology, 2019

674 citations recorded by the source corpus.

In recent years, object detection has experienced impressive progress. Despite these improvements, there is still a significant gap in the performance between the detection of small and large objects. We analyze the current state-of-the-art model, Mask-RCNN, on a challenging dataset, MS COCO. We show that the overlap between small ground-truth objects and the predicted anchors is much lower than the expected IoU threshold. We conjecture this is due to two factors; (1) only a few images are containing small objects, and (2) small objects do not appear enough even within each image containing them. We thus propose to oversample those images with small objects and augment each of those images by copy-pasting small objects many times. It allows us to trade off the quality of the detector on large objects with that on small objects. We evaluate different pasting augmentation strategies, and ultimately, we achieve 9.7\% relative improvement on the instance segmentation and 7.1\% on the object detection of small objects, compared to the current state of the art method on MS COCO.

Subfields: Computer Vision and Pattern Recognition, Radiology, Nuclear Medicine and Imaging

View Paper: https://www.semanticscholar.org/paper/8985726969b7ef4d19b6e28580efefefeb9d64be

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Radiology, Nuclear Medicine and Imaging

Citations in the corpus

QueryDet: Cascaded Sparse Query for Accelerating High-Resolution Small Object Detection (2021, methodology): “These methods can be mainly categorized into four types: 1) increasing the resolution of input features [31, 2, 23, 11, 28, 46, 26, 53]; 2) oversampling and strong data augmentation [31, 21, 75]; 3) incorporating context information [7, 64, 5], and 4) scale-aware training [28, 47, 48, 27].”
A Survey of the Four Pillars for Small Object Detection: Multiscale Representation, Contextual Information, Super-Resolution, and Region Proposal (2022, methodology): “images containing small objects and small object augmentation were also introduced to make the model focus more on the small objects [97].”
SCRDet++: Detecting Small, Cluttered and Rotated Objects via Instance-Level Feature Denoising and Rotation Loss Smoothing (2020, background): “Common small object solutions include data augmentation [49], multi-scale feature fusion [14], [50], tailored sampling strategies [30], [51], [52], generative adversarial networks [53], and multi-scale training [54] etc.”
RODEO: Replay for Online Object Detection (2020, neutral): “Recently, several augmentation strategies have been designed explicitly for object detection [28, 54, 60] and it would be interesting to explore how they could improve performance within deep feature space for an incremental learning application.”
Detecting tiny objects in aerial images: A normalized Wasserstein distance and a new benchmark (2022, neutral): “(Kisantal et al., 2019) propose to augment data by oversampling images containing tiny objects and copy-pasting tiny objects.”
Back to Reality: Weakly-supervised 3D Object Detection with Shape-guided Label Enhancement (2022, methodology): “As it is difficult for the network to extract highquality features of these objects, we utilize an augmentation strategy to alleviate the problem, which is similar to [14].”
Embracing Single Stride 3D Object Detector with Sparse Transformer (2021, methodology): “Some other methods adopt the scale-aware training [23, 25, 47] and strong data augmentations [18, 67].”
Slicing Aided Hyper Inference and Fine-Tuning for Small Object Detection (2022, methodology): “The method proposed in [16] oversamples images with small objects and augments them by making several copies of small objects.”
Towards Large-Scale Small Object Detection: Survey and Benchmarks (2022, methodology): “[55] adopted an augmentation strategy by copying a small object and pasting it with random transformation to different positions in the identical image.”