|
Jinsheng Wang I am a technical staff at Beijing Academy of Artificial Intelligence (BAAI), the vision and multimodal research center (a.k.a. BAAI Vision). I received my M.S. in EECS from Peking University. My research interests focus on multimodal foundation models and efficient scaling technologies. |
|
Selected ProjectsCore contributions as (co-)first author. Emu3: Next-Token Prediction is All You NeedEmu3-Team Breakthrough in multimodal scaling laws and native unified generation/understanding across text, image, and video modalities through next-token prediction. Loading... EVA-CLIP-18B: Scaling CLIP to 18 Billion ParametersQuan Sun*, Jinsheng Wang*, Qiying Yu*, Yufeng Cui, Fan Zhang, Xiaosong Zhang, Xinlong Wang The largest open-source CLIP model demonstrating efficient weak-to-strong scaling methodologies, achieving exceptional 80.7% zero-shot top-1 accuracy averaged across 27 widely recognized image classification benchmarks with optimized training strategies. Loading... Uni3D: Exploring Unified 3D Representation at ScaleJunsheng Zhou*, Jinsheng Wang*, Baorui Ma*, Yu-Shen Liu, Tiejun Huang, Xinlong Wang ICLR 2024 (Spotlight) - Pioneering scaling roadmap for 3D foundation models, establishing the first billion-parameter 3D understanding framework. Loading... GANet: A Keypoint-based Global Association Network for Lane DetectionJinsheng Wang*, Yinchao Ma*, Shaofei Huang*, Tianrui Hui, Fei Wang, Chen Qian, Tianzhu Zhang CVPR 2022 - Efficient lane detection achieving SOTA accuracy and high FPS. Successfully deployed in production with automotive industry partners (Bosch, Horizon Robotics). Demo available with YOLO integration. Loading... |
Experience
|
Selected Awards
|
MiscUnforgettable experiences in robotics competitions, emphasizing teamwork and pursuit of excellence. 
|