DeepGlint has 6 papers selected for ICCV 2025, covering areas such as visual foundation models and 3D face reconstruction

36Kr learned that recently, the international top conference ICCV 2025 (International Conference on Computer Vision) announced the results of paper acceptance, and DeepGlint had a total of 6 papers selected. The results involve several key directions in visual AI research, including visual foundation models, 3D face reconstruction, document geometric correction, human action generation, and robust 3D reconstruction. In the area of visual foundation models, a paper co-authored by DeepGlint and Huawei Noah's Ark Lab was accepted. This achievement effectively enhances the perception and expression capabilities of visual models for local area information by introducing regional Transformer layers and efficient regional clustering discriminative loss, making it perform outstandingly in dense visual tasks such as OCR, object detection, and segmentation