Chang D. Yoo

It is revealed that 2D image and 3D shape encoders exhibit largely disjoint failure patterns and rarely share identical wrong labels, whereas two 2D image encoders frequently repeat the same errors, which makes the 2D and 3D pair inherently complementary.

TanGO achieves state-of-the-art 3D editing by allowing fine-grained control over individual tokens, drastically reducing semantic artifacts.

Novel-view fidelity in drawing-based 3D animation can be drastically improved with a lightweight module that outperforms traditional fine-tuning methods.

GADA corrects spatial misalignments in Gaussian Splatting, preserving high-frequency details while achieving over twice the processing speed of existing techniques.

RTFree-F5 achieves a remarkable 10.4% WER on dysarthric speech without needing any reference transcripts, surpassing even ground-truth baselines.

VOTP slashes the labeling burden in preference-based reinforcement learning, achieving superior performance with minimal human input.

Single-step action generation can outperform multi-step diffusion methods in offline reinforcement learning, achieving higher performance with lower computational costs.

Diffusion Transformers waste up to 66% of their conditional embedding space without sacrificing generation quality, hinting at opportunities for more efficient conditioning.

Unlock the potential of your offline RL data: a new framework achieves state-of-the-art performance on D4RL benchmarks by quantifying and leveraging data uncertainty with a computationally efficient Rank-One MIMO architecture.