STAR-R1 enhances spatial reasoning in multimodal LLMs using reinforcement learning, achieving a 23% improvement in cross-view scenarios.

Zongzhao Li, Zongyang Ma, Mingze Li et al.

BAGEL employs large-scale interleaved multimodal pretraining with Mixture-of-Transformers, achieving emergent reasoning capabilities surpassing open-source models.

Chaorui Deng, Deyao Zhu, Kunchang Li et al.

CAD-Coder, an open-source VLM based on LLaVA fine-tuned for direct CAD code generation, achieves 100% syntax validity and high geometric accuracy.

Anna C. Doris, Md Ferdous Alam, Amin Heyrani Nobari et al.

Proposes CP-GS framework combining pre-trained image-to-3D models and LoRA fine-tuning to achieve multi-view consistent 3D scene personalization from a single image, outperforming SOTA.

Yuxuan Wang, Xuanyu Yi, Qingshan Xu et al.

Sparc3D combines sparse deformable Marching Cubes and a novel sparse convolutional VAE for high-res 3D shape reconstruction.

Zhihao Li, Yufei Wang, Heliang Zheng et al.

UniVG-R1 employs reasoning and reinforcement learning, achieving 9.1% performance gain on universal visual grounding tasks.

Sule Bai, Mingxing Li, Yong Liu et al.

Proposes Pensieve's two-stage training for high-quality novel view synthesis without calibration, achieving PSNR 26.53 on RealEstate10K.

Ruoyu Wang, Yi Ma, Shenghua Gao

CoSeP uses class-separability space clustering to automatically determine layer-wise pruning ratios, improving efficiency without sacrificing accuracy.

David Levin, Gonen Singer

PI-VAD enhances RGB features with multi-modal inputs, achieving state-of-the-art accuracy in video anomaly detection across three datasets.

Snehashis Majhi, Giacomo D'Amicantonio, Antitza Dantcheva et al.

BusterX leverages MLLM and reinforcement learning to detect AI-generated videos, utilizing a 200K high-quality dataset and producing interpretable reasoning chains, significantly improving accuracy and explainability.

Haiquan Wen, Yiwei He, Zhenglin Huang et al.

LogicOCR evaluates large multimodal models' logical reasoning on text-rich images using automated image generation and cropping, achieving 73% success rate.

Maoyuan Ye, Haibin He, Qihuang Zhong et al.

ContextAR model achieves efficient multi-conditional image generation, surpassing diffusion models.

Yixiao Chen, Zhiyuan Ma, Guoli Jia et al.

SpatialCrafter uses video diffusion models to reconstruct 3D scenes from sparse views, enhancing reconstruction accuracy.

Songchun Zhang, Huiyao Xu, Sitong Guo et al.

EgoDex leverages Apple Vision Pro to collect 829 hours of egocentric manipulation videos, enabling advanced imitation learning for dexterous robots.

Ryan Hoque, Peide Huang, David J. Yoon et al.

DPSeg integrates dual prompts and cost volume learning, significantly improving open-vocabulary semantic segmentation accuracy.

Ziyu Zhao, Xiaoguang Li, Linjia Shi et al.

Proposes Prior Depth Anything, integrating sparse measurements and depth prediction to produce dense, metric depth maps with strong zero-shot generalization.

Zehan Wang, Siyu Chen, Lihe Yang et al.

MathCoder-VL uses executable code for vision–math alignment, reaching 73.6% on MathVista GPS.

Ke Wang, Junting Pan, Linda Wei et al.

MIRAGE benchmark evaluates models' spatial perception in Counting, Relation, and Counting with Relation tasks.

Chonghan Liu, Haoran Wang, Felix Henry et al.

ResULIC combines Semantic Residual Coding and Compression-aware Diffusion for ultra-low bitrate image compression, saving over 66% in FID and LPIPS BD-rate.

Anle Ke, Xu Zhang, Tong Chen et al.

This paper introduces DanceGRPO, leveraging stable Group Relative Policy Optimization (GRPO) to significantly improve reinforcement learning for visual content generation, with up to 181% performance gains on benchmarks.

Zeyue Xue, Jie Wu, Yu Gao et al.