This paper proposes a precise scaling law for video diffusion Transformers, integrating optimal hyperparameter prediction to enhance performance and reduce inference costs by 40.1%.

Yuanyang Yin, Yaqi Zhao, Mingwu Zheng et al.

EPS uses DCT-based spatial-temporal features for efficient patch sampling, reducing training data by up to 91.69% with 82.1× speedup.

Yiying Wei, Hadi Amirpour, Jong Hwan Ko et al.

Zoom Eye enhances multimodal LLMs' visual reasoning with tree search, boosting InternVL2.5-8B by 15.71% on HR-Bench.

Haozhan Shen, Kangjia Zhao, Tiancheng Zhao et al.

Introduces orthogonal subspace decomposition via SVD to enhance generalization in AI-generated image detection, outperforming SOTA with >94% AUC.

Zhiyuan Yan, Jiangming Wang, Peng Jin et al.

Using k-SAE to interpret diffusion model features reveals hierarchical semantic information, with intermediate layers capturing fine details.

Dahye Kim, Xavier Thomas, Deepti Ghadiyaram

Proposes truncated diffusion with multi-mode anchors, reducing steps from 20 to 2, achieving 88.1 PDMS at 45FPS for autonomous driving.

Bencheng Liao, Shaoyu Chen, Haoran Yin et al.

OminiControl leverages parameter reuse of DiT's VAE and Transformer, adding only 0.1% parameters, to achieve versatile multi-task image control surpassing specialized methods.

Zhenxiong Tan, Songhua Liu, Xingyi Yang et al.

Stable Flow finds vital layers in DiT to enable training-free, stable image editing.

Omri Avrahami, Or Patashnik, Ohad Fried et al.

DINO-X achieves state-of-the-art open-world object detection with 56.0 AP on COCO.

Tianhe Ren, Yihao Chen, Qing Jiang et al.

EAST combines dynamic ReLU, weight sharing, and cyclic sparsity to achieve 99.99% sparsity with competitive accuracy.

Andy Li, Aiden Durrant, Milan Markovic et al.

Video-RAG integrates external visual-aligned texts via retrieval, no training needed, boosting long video comprehension by 2.8% on average.

Yongdong Luo, Xiawu Zheng, Guilin Li et al.

USP-Gaussian unifies spike-based image reconstruction, pose correction, and Gaussian splatting for enhanced 3D reconstruction accuracy.

Kang Chen, Jiyuan Zhang, Zecheng Hao et al.

EgoVid-5M provides 5M curated clips, while EgoDreamer jointly controls egocentric generation with text and kinematics.

Xiaofeng Wang, Kang Zhao, Feng Liu et al.

GaussianAnything employs point cloud structured latent space with diffusion models for multi-modal 3D generation, enabling high-quality, editable outputs.

Yushi Lan, Shangchen Zhou, Zhaoyang Lyu et al.

Proposes Image2Text2Image framework using diffusion models for label-free image caption evaluation, achieving high correlation with human judgment.

Jia-Hong Huang, Hongyi Zhu, Yixian Shen et al.

Multi-modal self-supervised learning enhances CVD prediction, improving balanced accuracy by 7.6% on limited data using ECG, CMR, and clinical info.

Francesco Girlanda, Olga Demler, Bjoern Menze et al.

SG-I2V enables zero-shot trajectory control in image-to-video generation using pre-trained diffusion models, achieving high-quality, controllable videos without fine-tuning.

Koichi Namekata, Sherwin Bahmani, Ziyi Wu et al.

Proposes a recurrent-based variable-length image tokenizer, adapting token count (32-256) to image complexity for efficient compression.

Shivam Duggal, Phillip Isola, Antonio Torralba et al.

EVSNet leverages event-based motion features with a lightweight fusion framework, achieving 34.1% mIoU on low-light VSPW, 11× parameter efficiency over SOTA.

Zhen Yao, Mooi Choo Chuah

NoPoSplat reconstructs 3D Gaussian scenes from sparse unposed images in real-time, surpassing pose-required methods.

Botao Ye, Sifei Liu, Haofei Xu et al.