European Conference on Computer Vision (ECCV) was held online on August 23-28. School of Artificial Intelligence of Xidian University organized graduate students to participate in competitions held at this conference, winning 1 champion, 5 second places, and 1 third place.

In the challenge of Vision Meets Drone: A Challenge that ended on July 15th, two joint teams formed by Xidian and WeBank won the championship and runner-up in the single target tracking competition respectively. This competition was aims at the scenes shot from the UAV’s perspective. It needs to deal with the tracking problems of similar object interference, occlusion, target deformation, night vision and other scenes, which posed a great greater challenge to the robustness of the tracking algorithm.

In the challenge ofUnder-Display Cameraon July 26, two teams won prizes. One team won the runner-up in the first track, and the other team won the fifth place in the second track. This competition requires to restore pictures taken by the camera under the screen. Due to the light transmittance and diffraction effects of the screen, the camera images under the screen were often of low quality, showing a variety of complex combined degradation types (such as noise, blur, etc.), and it was quite difficult to restore them.

In the challenge of Drama QA on July 31, Xidian team won the runner-up. This competition requires algorithms to be able to answer the correct options corresponding to the question in a given video sequence and visual annotations (such as characters, places, objects and objects related to characters). This task involved multi-modal information processing of images and texts, and needed to evaluate the accuracy of the selected answer in the test set.

In the challenge of Giga Vision 2020 on August 15th, teams won the second place,the fourth place andthe fifth placein the target detection track, therunner-up,the third place,the fourth placein the multi-target detection track. This competition requires target detection and multi-target tracking on billion-pixel-level images and videos. Target detection needs to predict the position of the person’s whole body, visible part of body, head, as well as the position of strollers, bicycles in a given test set. Multi-target tracking needs to track everyone in the scene. The data set had large images, many targets, low frame rate (2 frames per second), and problems such as complex occlusion.

At the same time, many papers written by students of School of Artificial Intelligence were accepted by the conference ECCV 2020s.
In recent years, students of School of Artificial Intelligence have achieved excellent results in many international top-level competitions.