Our paper on Open Domain Generalization for Vision-Language Models, as one of the world’s first, to recognize samples from out-of-distribution classes of out-of-distribution domains, using an unknown text prompt only, by adopting a diffusion-based unknown data generation strategy and incorporating text features into vision, has been accepted in CVPR 2024, a premier international conference for research and applications of computer vision. In the future, we aim to unleash the potential of vision-language conjuncture in autonomous driving for robust recognition of out-of-distribution domains in a diverse environment with the use of human-language interference.

The IEEE/CVF Conference on Computer Vision and Pattern Recognition 2024
Unknown Prompt, the only Lacuna: Unveiling CLIP’s Potential for Open Domain Generalization

Mainak Singha (Tokyo Research Center, Aisin Corporation)
Ankit Jha (Indian Institute of Technology Bombay)
Shirsha Bose (Technische Universität München)
Ashwin Nair (Indian Institute of Science Education and Research Thiruvananthapuram)
Moloud Abdar (Deakin University)
Biplab Banerjee (Indian Institute of Technology Bombay)

関連リンク：
https://cvpr.thecvf.com/Conferences/2024/AcceptedPapers