🏢 Tsinghua University

MeshRet: A novel retargeting framework that uses dense geometric interaction modeling for realistic, artifact-free skinned character animation.

ShowMaker: Generating high-fidelity 2D human conversational videos using fine-grained diffusion modeling and 2D key points.

SG-Nav achieves state-of-the-art zero-shot object navigation by leveraging a novel 3D scene graph to provide rich context for LLM-based reasoning.

HERT: a novel framework for semi-open 3D object retrieval using hierarchical hypergraph equilibrium, achieving state-of-the-art performance on four new benchmark datasets.

Unlocking the potential of deep learning for time series forecasting: this study reveals a scaling law influenced by dataset size, model complexity, and the crucial look-back horizon, leading to impro…

FOCALBO, a hierarchical Bayesian optimization algorithm using focalized sparse Gaussian processes, efficiently tackles high-dimensional problems with massive datasets, achieving state-of-the-art perfo…

S-STE achieves efficient 2:4 sparse pre-training by introducing a novel continuous pruning function, overcoming the limitations of previous methods and leading to improved accuracy and speed.

ROPINN: Revolutionizing Physics-Informed Neural Networks with Region Optimization

RGB-Event tracker CSAM leverages MOT philosophy for enhanced robustness by integrating appearance and motion information from RGB and event streams, achieving state-of-the-art performance.

ReST-MCTS*: A novel LLM self-training method using process reward guided tree search, outperforming existing methods by generating higher-quality reasoning traces for improved model accuracy.

AI-powered dual-target drug design is revolutionized by repurposing pretrained diffusion models, achieving zero-shot transfer learning and outperforming existing methods.

ReFIR enhances Large Restoration Models’ accuracy by incorporating retrieved images as external knowledge, mitigating hallucination without retraining.

Boost skeleton action recognition accuracy across datasets by recovering complete actions and resampling; outperforms existing methods.

RealCompo: A novel training-free framework dynamically balances realism and compositionality in text-to-image generation, achieving state-of-the-art results.

CORUN-Colabator: a novel cooperative unfolding network and coherence-based label generator achieves state-of-the-art real-world image dehazing by effectively integrating physical knowledge and generat…

Q-VLM: A novel post-training quantization framework significantly compresses large vision-language models, boosting inference speed without sacrificing accuracy.

PAD, a novel visual policy learning framework, unifies image prediction and robot action in a joint denoising process, achieving significant performance improvements in robotic manipulation tasks.

PHYRECON: A novel neural scene reconstruction method uses differentiable rendering and physics simulation for physically plausible 3D models.

PEAC: a novel unsupervised pre-training method significantly improves cross-embodiment generalization in reinforcement learning, enabling faster adaptation to diverse robots and tasks.

Parameter-Inverted Image Pyramid Networks (PIIP) boost image pyramid efficiency by using smaller models for higher-resolution images and larger models for lower-resolution ones, achieving superior per…