WORKSHOP ON "Deep learning based small object detection from images and videos"

Object detection has become an inseparable part of many modern high-level computer vision tasks. Modern deep CNNs and Transformer-based networks (e.g., YOLO, Faster R-CNN, ViT, DETR) have shown great success and performance in identifying and localizing objects in highly complex images or videos. Despite the huge improvements of deep learning methods over classical object detection frameworks, they still fail to accurately detect small objects which only cover a small portion of the images or video frames. There exists a huge gap in the performance of State-Of-The-Art (SOTA) object detection techniques when facing normal and small objects. This poor performance is due to the lack of geometrical information in deeper layers of the network, inappropriate evaluation metrics, a limited number of small objects in datasets, noisy labelled datasets, etc. Although the standard deep learning-based methods have adequate power to be used in many general real applications, they need to be extended to be suitable for applications such as pedestrian detection and sign detection in autonomous driving, tumor detection in medical images, industrial product quality assessment from a visual perspective, face recognition, and ship and swimmer detection in the maritime environment. Above applications mostly deal with optical images and videos (RGB, LWIR, NIR, IR, Hyperspectral) acquired from satellites, cars, ships, UAVs, USVs, etc.

Thus motivated, we organize the ACCV 2022 workshop on “Deep Learning-Based Small Object Detection from Images and Videos” in Dec 2022, in conjunction with Asia Conferences on Computer Vision (ACCV 2022) in Macau, China. we invite researchers to submit their latest research output to this workshop and present it in a poster/oral presentation. This workshop focuses on optical data modalities (RGB, multispectral, hyperspectral images or videos), regardless of their source (i.e., satellite or else). The theme of the workshop is not limited to small objects, but also small objects in challenging environments, e.g., the maritime environment in which Synthetic Aperture Radar (SAR) based methods are still the leading technology, small object detection in the presence of haze, and class-agnostic object detection, etc.