PRIS-CV: Computer Vision Group

Autonomous synthetic data construction via VLM-guided iterative rendering — let your data build and improve itself.

[PRCV 2026] Fine-grained benchmark and evaluation framework for text-to-video generation using ST-DAG prompt decomposition, VLM-based QA, and objective/subjective scoring.

[MICCAI 2026] The official code for "Hepato-LLaVA: An Expert MLLM with Sparse Topo-Pack Attention for Hepatocellular Pathology Analysis on Whole Slide Images"

We propose FlexiGrad, a simple and parameter-free method for hierarchical fine-grained classification that regulates gradient interactions during backpropagation.

[ECCV 2026] Code release for "Recolour What Matters: Region-Aware Colour Editing via Token-Level Diffusion"

[NeurIPS 2025] Self-Supervised Selective-Guided Diffusion Model for Old-Photo Face Restoration

[AAAI 2026] The official code for "MedReasoner: Reinforcement Learning Drives Reasoning Grounding from Clinical Thought to Pixel-Level Precision"

Official PyTorch implementation of **"Implicit Diffusion Distillation with Domain Disentanglement for Domain Generalization"** (ICME 2026).