TaskCLIP leverages large vision-language models with a transformer aligner to improve task-oriented object detection by 3.5%, requiring only one RTX4090.

Hanning Chen, Wenjun Huang, Yang Ni et al.

DriveDreamer-2 uses LLMs and a Unified Multi-View Model to generate customized driving videos; FID 11.2, FVD 55.7.

Guosheng Zhao, Xiaofeng Wang, Zheng Zhu et al.

FastV learns adaptive attention and prunes visual tokens, reducing FLOPs by up to 45% in LVLMs like LLaVA-1.5-13B, with minimal performance loss.

Liang Chen, Haozhe Zhao, Tianyu Liu et al.

VidProM is a large-scale dataset with 1.67M user prompts and 6.69M videos from 4 diffusion models, advancing text-to-video research.

Wenhao Wang, Yi Yang

ELLA integrates LLM with diffusion models using TSC for enhanced semantic alignment.

Xiwei Hu, Rui Wang, Yixiao Fang et al.

Proposed CAT model enhances multimodal reasoning with clue aggregation and GPT-based preference optimization, boosting AVQA performance.

Qilang Ye, Zitong Yu, Rui Shao et al.

TextMonkey: OCR-free large model, improves document understanding, scores 561 on OCRBench.

Yuliang Liu, Biao Yang, Qiang Liu et al.

This survey reviews controllable generation in text-to-image diffusion models, analyzing DDPMs' control mechanisms.

Pu Cao, Feng Zhou, Qing Song et al.

SceneCraft uses LLMs to convert text into Blender scripts, enabling the creation of complex scenes with up to 100 assets.

Ziniu Hu, Ahmet Iscen, Aashi Jain et al.

Proposes Selective Recurrent Unit (SRU) with adaptive frequency fusion, achieving top accuracy in stereo matching, ranked first on KITTI and other benchmarks.

Xianqi Wang, Gangwei Xu, Hao Jia et al.

TempCompass benchmark evaluates 8 SOTA Video LLMs across five temporal dimensions, revealing their poor temporal perception abilities, with average accuracy around 33.9%.

Yuanxin Liu, Shicheng Li, Yi Liu et al.

RAG combines retrieval with generation, boosting accuracy and robustness in AI-generated content.

Penghao Zhao, Hailin Zhang, Qinhan Yu et al.

Proposes a diffusion-based multimodal visual-audio generation framework using a shared latent aligner, significantly improving joint content quality.

Yazhou Xing, Yingqing He, Zeyue Tian et al.

Sora uses a diffusion transformer with spacetime patches to generate high-quality 60s videos from text.

Yixin Liu, Kai Zhang, Yuan Li et al.

NaVid is a video-based vision-language navigation model that uses monocular RGB input to achieve state-of-the-art performance without maps or depth data, demonstrating strong Sim2Real transfer.

Jiazhao Zhang, Kunyu Wang, Rongtao Xu et al.

Introduces PGI and GELAN to improve gradient reliability and parameter efficiency, boosting YOLOv9 detection performance.

Chien-Yao Wang, I-Hau Yeh, Hong-Yuan Mark Liao

VADv2 introduces a probabilistic scene-conditioned planning model for end-to-end autonomous driving, achieving state-of-the-art performance on CARLA benchmark with 85.1% success rate.

Bo Jiang, Shaoyu Chen, Hao Gao et al.

Proposed ChartX benchmark and ChartVLM model achieve near GPT-4V performance in complex chart reasoning.

Renqiu Xia, Bo Zhang, Hancheng Ye et al.

Proposes Any-Shift Prompting, a hierarchical Bayesian framework, achieving significant improvements over 23 datasets in cross-distribution generalization.

Zehao Xiao, Jiayi Shen, Mohammad Mahdi Derakhshani et al.

GALA3D employs LLM-generated layouts and Gaussian splatting for high-fidelity complex scene synthesis, outperforming prior methods with CLIP scores of 35.052.

Xiaoyu Zhou, Xingjian Ran, Yajiao Xiong et al.