The AI Vision Future team has been honored with the selection of two pioneering research projects for the prestigious ICCV 2025 conference. These projects, which concentrate on image generation and video enhancement, represent significant advancements in the field of visual technology.

In the realm of image generation, the team has introduced the De-MAR framework. This innovative approach leverages a dual-token optimization mechanism, combined with diffusion and denoising head modules, to achieve efficient image creation. Experimental results on the ImageNet and MS-COCO datasets reveal that De-MAR achieves impressive FID scores of 1.47 and 5.27, respectively, while generating images 45% faster than the DiT-XL/2 model.

Turning to video enhancement, the team's GenVE framework stands out for its ability to restore details with remarkable vividness, resulting in video frames that appear more natural and smooth. This breakthrough underscores the potential of AI in enhancing the quality of visual content.