Earth vision, also known as Earth Observation and Remote Sensing, targets to understand large-scale scenes on the Earth’s surface with aerial images taken from the overhead view, which provides a new way to understand our physical
world and benefits many applications, e.g., urban management and planning, precise agriculture, emergency rescue, and disaster relief. However, in contrast with nature images, aerial images have many distinct properties in the viewpoint
of understanding and model designs.
Aerial images are not human-centric while mainly for photogrammetry or measurement. Thus, the significant visual elements of an aerial scene may be distributed uniformly in images, and the compositional rule, if existing, behind aerial images is largely
different from that of nature images.
The geometries of aerial images are different from those of nature images. For instance, objects in aerial images often appear in arbitrary orientations, and the scale variations of object instances in aerial images are considerably huge. And small objects
may densely be distributed in aerial images.
Besides, the spectral bands, focus lengths, depth fields, and imaging geometries of aerial images are also different from those of nature images.
This workshop on
, aims to draw attention from a wide range of communities and call for more future research and efforts on the problems of object detection, instance segmentation, and semantic segmentation
in aerial images. The workshop also contains a competition with 3 tasks (object detection with oriented bounding boxes, object detection with horizontal bounding boxes, semantic segmentation) of this workshop.
Topics of interests include, but are not limited to following fields:
Oriented/tiny/crowded object detection in aerial images
Instance segmentation in aerial images
Semantic segmentation/classification in aerial images
Change detection/analysis in aerial image sequences
Object tracking in aerial image sequences
Deep models for aerial image understanding
Explainable models for aerial image understanding
Weakly-supervised/Unsupervised models for aerial image understanding
Benchmarks and datasets for aerial image understanding
Reviews and perspectives in Earth vision
Papers will be limited to 8 pages excluding references according to the ICCV’2021 format. (main conference authors’ guidelines). One can download the templates at
LaTex Templates(zip). We will assign each paper to at least 3 reviewers with a double-blind policy. Papers will be accepted according to relevance,
significance, technical novelty, and clarity of presentation. Papers will be published in ICCV’2021 workshop proceedings. Several papers will be selected as oral representation on the workshop. All the papers should be submitted through