A comprehensive index of artificial intelligence and machine-learning research with AI-generated summaries, citation metrics, and direct links to papers and code.
Unknown
SSDNeRF unifies 3D generation and reconstruction by conjoining a triplane NeRF auto-decoder with a single-stage diffusion model.
Unknown
Tune-A-Video enables one-shot text-to-video generation by tuning a pretrained image diffusion model on a single video.
Liangchen Li, Jiajun He
Bidirectional consistency models extend consistency models to enable both forward and reverse sampling, improving image generation quality and flexibility.
Unknown
AudioLCM introduces a latent consistency model for efficient text-to-audio generation, enabling high-quality synthesis in few sampling steps.
Unknown
This paper introduces flow matching and diffusion models, explaining and deriving flow matching as a scalable algorithm underlying large-scale generative models.
Unknown
Q-dit introduces a post-training quantization method for diffusion transformers that reduces memory and latency while maintaining generation quality.
Unknown
Lazydit introduces a lazy learning approach to accelerate Diffusion Transformers by selectively skipping redundant computations during inference.
Unknown
The paper argues that semantically richer latent representations can substantially improve diffusion Transformer optimization efficiency and expand the reconstruction–generation Pareto frontier.
Unknown
The paper shows that vision foundation models can serve as effective tokenizers for latent diffusion models, but distillation-based tokenizer training weakens performance.
Cheng Chi, Zhenjia Xu, Siyuan Feng, et al.
Diffusion Policy represents robot visuomotor policies as conditional denoising diffusion processes, outperforming state-of-the-art methods by 46.9% across 15 tasks.
Ying-Chao Ma, Anders Oland, Anton Ragni, et al.
This paper provides a comprehensive review of foundation models in music, covering representation, generative, and multimodal learning, and highlighting underexplored areas and future directions.
Chirag Vashist, Shichong Peng, Ke Li
RS-IMLE improves few-shot image synthesis by redesigning the prior distribution to align training and inference latent codes, yielding higher quality images than existing GAN and IMLE methods.