ziqihuangg

Professional software vendor delivering innovative solutions on the Softono platform. Specialized in both open-source and proprietary software development.

Visit Website

Total Products

Software by ziqihuangg

Open Source

ReVersion

ReVersion is a diffusion-based framework for a new computer vision task called Relation Inversion. Presented at SIGGRAPH Asia 2024 by researchers from MMLab at Nanyang Technological University, ReVersion allows users to discover and capture specific interactions or relations between entities from a few exemplar images. Given input images containing a consistent relationship, the system generates a compact relation prompt that encodes this interaction. Once learned, this prompt can be applied to entirely new entities to synthesize novel scenes where the extracted relationship is preserved. The software enables creative image generation by extending learned concepts like specific poses, collaborations, or physical interactions across different subjects. ReVersion features an optimized implementation where only the lightweight relation prompt needs to be saved and loaded, eliminating the need to store the full text-to-image model. The project includes a released benchmark dataset, pre-trained relation prompts, t

ML Frameworks

504 Github Stars

Open Source

Collaborative Diffusion

Collaborative Diffusion is a research framework presented at CVPR 2023 for multi-modal face generation and editing. Developed by ziqihuangg from MMLab at Nanyang Technological University, this software leverages pre-trained unimodal diffusion models to synthesize high-quality facial images guided by multiple input conditions such as text, attributes, or spatial maps. The core innovation is a dynamic diffuser mechanism that operates during the reverse diffusion process to calculate an influence function, allowing the system to selectively enhance or suppress specific modalities at each timestep. This approach ensures high consistency between the generated output and the diverse input controls. Key features include robust multi-modal control for creating new faces, precise editing capabilities for real images with strong identity preservation, and support for 512x512 and 256x256 resolutions. The repository provides full training and inference code, pre-trained checkpoints, and compatibility with FreeU for enhan

AI & Machine Learning ML Frameworks

441 Github Stars