European Conference on Computer Vision selects 60 SenseTime papers, face biometrics bias challenge held
SenseTime has announced that 60 papers authored or co-authored by its employees, several with implications for biometric technology, have been accepted to the European Conference on Computer Vision (ECCV).
The biennial ECCV is the continent’s top event for image analysis. The 2020 edition, the 16th, is taking place online this year due to the COVID-19 pandemic. ECCV 2020 includes the 2020 Looking at People Fair Face Recognition challenge, presented by ChaLearn. Participants will attempt to develop biometric facial verification methods for reduced disparities in accuracy between different genders and skin colors, and is sponsored by AnyVision.
There were 5,025 paper submitted to ECCV 2020, with only 1,361 accepted from among them.
SenseTime says it also had high results in the Large Vocabulary Instance Segmentation (LVIS), which Facebook AI Research introduced in 2018, and the Advances in Image Manipulation workshop (AIM 2020) at ECCV 2020.
The papers from SenseTime address computer vision challenges including generative adversarial networks (GANs), 3D point cloud understanding, and object recognition. One example is a paper describing the use of a GAN to restore missing semantics in various degraded images, and another focusses on the use of RobustScanner for dynamic enhancement of positional clues for robust test recognition.
By applying a two-stage fine-tuning method to improving its algorithms and get beyond class imbalance problems, SenseTime says it was able to improve its segmentation mask detection for more efficient and accurate long-tail object recognition.
SenseTime has been doing brisk trade in its domestic Chinese market, and the company’s revenue is reported to have grown by 147 percent in 2019.