Tsinghua AI

Human-crafted conference posters are now directly usable as verified supervision for generating new posters, bypassing the limitations of layout-box-only datasets.

Packed trajectories, not sequential rollouts, unlock high-fidelity multi-turn image editing by preventing visual drift.

A single scene can be represented by ~100K plane primitives (5MB), enabling video world models to maintain geometric consistency across hundreds of frames and large viewpoint changes.

Visual generation supervision injects significant, yet often inaccessible, knowledge into UMMs, requiring novel alignment strategies to unlock its full potential for text-based reasoning.

Training-free conditional sampling for Schrödinger Bridges is now possible, achieving state-of-the-art results on image generation and inverse problems by twisting SB's inherent endpoint prediction capabilities.

Ultralight action heads achieve state-of-the-art embodied task performance by learning to selectively retrieve visual information via shared latent bridge queries.

A single regional weather model can now leverage multiple global forecast drivers without retraining, achieving kilometer-scale accuracy.

Ambiguous object selection in text-guided image editing is solved by directly annotating the canvas, achieving state-of-the-art performance in complex edits.

Decoupling denoising and context writing parameters in autoregressive video models unlocks unprecedented long-horizon generation capabilities, extending continuous output to 24 hours from just 5-second training rollouts.

A 260M-parameter looped diffusion model achieves performance surpassing a 6.5x larger non-looped counterpart with significantly less inference compute.

Video agents can now autonomously discover and fix their own reasoning flaws by re-watching videos and evolving their evidence-gathering strategies.

Single-step diffusion models can outperform their multi-step teachers when distributional matching is reformulated as a Wasserstein gradient flow over Gaussian mixtures.

Diffusion teachers and online score networks are no longer bottlenecks for distilling autoregressive video models: directly minimizing sample MMD against reference videos improves VBench scores while unlocking 14B post-training on just eight H200 GPUs.

DART, a training-free method that combines low-rank coordinate transport with target-schedule response calibration with forward evaluations and no source training videos is proposed, and Adapter-level results reveal positive functional effects for some adapters and strong attenuation with reduced negative functional effects for others.

Vidu S2, which comprises Vidu S2-Avatar, a real-time interactive digital-character model, and Vidu S2-Editing, a real-time video editing model, is presented and the feasibility of real-time spatial video generation for both Vidu S2-Avatar and Vidu S2-Editing is explored.

Neural image compression can finally rival classic formats in raw throughput, achieving 2000 FPS decoding and practical 20 FPS encoding while matching JPEG's rate-distortion performance.

A controlled pure-autoregressive testbed is built and track task-specific validation losses during multimodal continual pretraining across text, image, text-to-image (T2I), and image-to-text (I2T) prediction, showing that better reconstruction does not necessarily yield lower task-specific losses or stronger downstream performance, and that image tokenizer choice can affect text modeling under joint optimization.

Single vision foundation models inevitably overfit to deepfake artifacts, but dynamically arbitrating between CLIP's semantics and DINO's structural features via prediction entropy delivers state-of-the-art generalization across unseen generators.

ToPO achieves superior image generation performance by optimizing token-conditioned preferences, outperforming existing methods across multiple evaluation metrics.

Visual design no longer requires choosing between diffusion fidelity and code editability: orchestrating modular asset diffusion through a VLM-driven HTML/CSS coding loop delivers fully interactive, layer-decoupled graphic layouts.

Information density imbalance, not just instance count, is a key driver of category bias in object detection models, revealing new avenues for enhancing fairness and accuracy.

SR-Edit achieves unprecedented image editing fidelity by refining region separation and preserving non-edit areas without introducing new artifacts.

Unpaired data can be leveraged to achieve state-of-the-art performance in solving complex inverse problems, challenging the reliance on paired datasets.

Current neuron reconstruction methods fall short at the whole-brain scale, revealing critical gaps in structural accuracy that CORAL aims to address.

Object-Uni transforms how we understand and generate spatial representations of objects, enabling precise manipulation of their poses in generated images.

Emotion-aware art generation just got a major upgrade with ReART, achieving perfect alignment with target emotions through structured visual retrieval.

By combining images with text queries, ID-VTG significantly improves the accuracy of video grounding in scenarios with visually similar entities.

MSEditor achieves unprecedented consistency in multi-shot video editing, outperforming existing methods by effectively managing identity drift and temporal coherence.

Separating spatial and temporal modeling in action recognition leads to significant performance gains, challenging the effectiveness of existing implicit coupling methods.

GeoWeaver achieves unprecedented accuracy in long-sequence 3D reconstruction by correcting accumulated errors through a novel combination of geometric priors and adaptive refinement techniques.

Achieving the best macro-average results on dynamic scene reconstruction benchmarks, UniQuery4R redefines efficiency in 4D scene understanding.

GS-Voxel revolutionizes 3D scene generation by allowing scalable, fitting-free structured latents that adapt to the complexity of the scene without the need for per-scene optimization.

StreamOPD achieves near teacher-level performance in streaming video understanding without relying on memory or retrieval, reshaping the landscape of post-training techniques.

Achieving 66.9 MOTA in 3D tracking without LiDAR reveals a new frontier in roadside infrastructure understanding.

Compositional zero-shot singing-and-dancing generation is now possible, allowing for unprecedented flexibility in video creation from audio and visual prompts.

Vision-based tactile sensors could revolutionize robotic interaction by providing high-resolution tactile data that enhances perception and manipulation capabilities.

Detecting partially forged videos is now feasible with a novel framework that leverages static images for enhanced supervision and accuracy.

HumanScore reveals that traditional kinematic metrics overlook critical failures in humanoid motion tracking, such as unstable support and incorrect contacts.

Verifiable temporal grounding in video forensics can drastically improve the detection of AI-generated content, outperforming traditional methods reliant on coarse supervision.

Pruning 50% of channels in RGB-infrared object detectors can actually boost performance by 0.6% mAP, challenging conventional wisdom about redundancy.

Bypassing RGB entirely, Latent-to-4D achieves significant improvements in 4D scene generation while maintaining a reusable framework across different video models.

Cross-frame feedback boosts Transformer tracking performance by leveraging historical information, outperforming traditional same-frame methods by up to 3.2 AO points.

Achieving state-of-the-art performance in both 4D reconstruction and point tracking, Uni4R leverages continuous velocity fields to model dynamics at any timestamp, breaking free from traditional limitations.

Unpaired MRI translation can effectively preserve anatomical structures across varying magnetic field strengths, a significant leap in addressing domain shift challenges in medical imaging.

GameAlpha-2.4K not only revolutionizes RGBA video generation for gaming but also achieves a significant efficiency boost by intelligently bypassing redundant computations.

A multi-agent forensic reasoning framework outperforms leading closed-source models in deepfake detection by leveraging diverse analytical perspectives on forgery cues.

EffectLearner achieves unprecedented video object removal quality by effectively reasoning about complex object-induced effects in dynamic scenes.