Seminar

Zoom Link  https://hku.zoom.us/meetings/95832932074/invitations?signature=SCg4igomh110T7WUpW6op2xeFgb9WgU92xDTYI2sp8s Abstract Text-to-image generation models have achieved impressive visual quality, yet they largely fail when prompts require reasoning or world knowledge to interpret. Generating an image of "a park thirty minutes after heavy rainfall" or "the city hosting the 2021 Summer Olympics" demands inference and domain knowledge, not just pattern matching. This seminar […]

Zoom Link  https://hku.zoom.us/s/92652052314?pwd=kKKw9mhMHxP8xcObYHFop7VyvAfzh1.1#success Abstract Autoregressive (AR) models have emerged as a powerful paradigm for text-to-image and text-to-video generation, exhibiting exceptional native in-context learning capabilities and scalable performance. However, their real-world deployment remains severely bottlenecked by the conventional sequential, token-by-token decoding process, which incurs immense inference latency by demanding thousands of sequential forward passes. To break […]

Zoom Link  https://hku.zoom.us/j/94632758376 Abstract Additive manufacturing enables customised device fabrication for emerging sensing technologies.  However, printable (opto)electronic devices with sophisticated architectures, including all-printed photodiodes, face challenges in multi-material and multi-layer printing at micro- and nanoscales with low processing temperatures.  Herein, we establish a nano-resolution printing method based on electrohydrodynamic printing (EHDP) to deposit inks from […]

Zoom Link  https://hku.zoom.us/j/3623268282 Abstract Creating high-quality 3D content requires not only visual realism but also controllability, editability, and structural understanding. This talk focuses on recent progress toward controllable 3D content generation, tracing a path from holistic object-level generation to part-aware modeling. It begins with object-level controllable 3D generation enabled by multi-view conditioning, aiming to guide […]