Computer vision

Anomaly detection plays a critical role in various industries for quality control and safety monitoring. The common methods of anomaly detection involve using self-supervised feature reconstruction. However, these techniques are often challenged by the need to create diverse and realistic anomaly samples while reducing feature redundancy and eliminating pre-training bias. Researchers from the College of Information…

Japanese comics, known as Manga, have gained worldwide admiration for their intricate plots and unique artistic style. However, a critical segment of potential readers remains largely underserved: individuals with visual impairments, who often cannot engage with the stories, characters, and worlds created by Manga artists due to their visual-centric nature. Current solutions primarily rely on…

The intersection of Artificial Intelligence's (AI) language understanding and visual perception is evolving rapidly, pushing the boundaries of machine interpretation and interactivity. A group of researchers from the Korea Advanced Institute of Science and Technology (KAIST) has stepped forward with a significant contribution in this dynamic area, a model named MoAI. MoAI represents a new…

Text-to-video diffusion models are revolutionizing how individuals generate and interact with media. These advanced algorithms can produce engaging, high-definition videos just by using basic text descriptions, enabling the creation of scenes that vary from serene, picturesque landscapes to wild and imaginative scenarios. However, until now, the field's progress has been hindered by a lack of…

Subject-driven image generation has seen a remarkable evolution, thanks to researchers from Alibaba Group, Peking University, Tsinghua University, and Pengcheng Laboratory. Their new cutting-edge approach, known as Subject-Derived Regularization (SuDe), improves how images are created from text-based descriptions by offering an intricately nuanced model that captures the specific attributes of the subject while incorporating its…

In the world of artificial intelligence (AI), integrating vision and language has been a longstanding challenge. A new research paper introduces Strongly Supervised pre-training with ScreenShots (S4), a new method that harnesses the power of vision-language models (VLMs) using the extensive data available from web screenshots. By bridging the gap between traditional pre-training paradigms and…