We propose CoT-RVS to extract the temporal-semantic correlation in videos with chain of thoughts and achieve the state-of-the-art performance for reasoning video segmentation.

We propose ThinkFirst, a novel Chain-of-Thought (CoT) reasoning segmentation framework that generates an accurate object mask given a text prompt. ThinkFirst can handle difficult scenarios such as implicit queries, camouflaged objects, out-of-domain objects with easy control.

We introduce UVRM, a novel 3D reconstruction model capable of being trained and evaluated on 360-degree monocular videos without requiring any information about the pose.

We propose StreamGS, an online generalizable 3DGS reconstruction method for unposed image streams, which progressively transform image streams to 3D Gaussian streams by predicting and aggregating per-frame Gaussians.

We propose invertible Guided Consistency Training (iGCT), a data-driven training framework for guided consistency model, contributing to fast and guided image generation and editing without requiring the training and distillation of any diffusion models.

We propose InceptionHuman, a NeRF-based generative framework incorporating state-of-the-art diffusion models, which receives any types and any sizes of prompts, (e.g. text, pose, style) to generate realistic 3D human.

We introduce Deceptive-NeRF/3DGS, a new method for enhancing the quality of reconstructed NeRF/3DGS models using synthetically generated pseudo-observations, capable of handling sparse input and removing floater artifacts.

We reveal the issue of Inter-block Optimization Entanglement (IBOE) in end-to-end KD training and further propose StableKD to stablilize optimization. Extensive experiments show StableKD achieve high accuracy, fast convergence, and high data efficiency.

We propose a simple yet fast and effective partial convolution (PConv), as well as a latency-efficient family of network architectures called FasterNet.

Conduct cutting-edge research and publish state-of-the-art papers in the 3D computer vision domain, contributing significantly to advancements in the research community.

Design and develop algorithms to solve problems; researching the latest technologies; tasks including image processing, OCR, data parsing, deep learning, etc. [report]