1

Imaginative Perception Tokens Enhance Spatial Reasoning in Multimodal Language Models
Imaginative Perception Tokens enhance spatial reasoning in multimodal language models - CVPR 2026 MUSI Workshop
Mahtab Bigverdi, Linjie Li, Weikai Huang, Yiming Liu, Jaemin Cho, Jieyu Zhang, Tuhin Kundu, Chris Dongjoo Kim, Zelun Luo, Ranjay Krishna, Linda Shapiro
Cite
One Life to Learn: Inferring Symbolic World Models for Stochastic Environments from Unguided Exploration
OneLife learns symbolic programmatic world models from a single episode of unguided exploration in stochastic environments
Zaid Khan, Archiki Prasad, Elias Stengel-Eskin, Jaemin Cho, Mohit Bansal
Preprint Cite Project
One Life to Learn: Inferring Symbolic World Models for Stochastic Environments from Unguided Exploration
RotBench: Evaluating Multimodal Large Language Models on Identifying Image Rotation
a benchmark evaluating MLLMs’ ability to identify image rotation
Tianyi Niu, Jaemin Cho, Elias Stengel-Eskin, Mohit Bansal
Preprint Cite Code
RotBench: Evaluating Multimodal Large Language Models on Identifying Image Rotation
Bifrost-1: Bridging Multimodal LLMs and Diffusion Models with Patch-level CLIP Latents
a unified framework that bridges multimodal LLMs and diffusion models with patch-level CLIP latents
Han Lin, Jaemin Cho, Amir Zadeh, Chuan Li, Mohit Bansal
Preprint Cite Code Project
Bifrost-1: Bridging Multimodal LLMs and Diffusion Models with Patch-level CLIP Latents
Video-Skill-CoT: Skill-based Chain-of-Thoughts for Domain-Adaptive Video Reasoning
a framework that automatically constructs and leverages skill-aware CoT supervisions for domain-adaptive video reasoning
Daeun Lee*, Jaehong Yoon*, Jaemin Cho, Mohit Bansal
Preprint Cite Code Project
Video-Skill-CoT: Skill-based Chain-of-Thoughts for Domain-Adaptive Video Reasoning
CAPTURe: Evaluating Spatial Reasoning in Vision Language Models via Occluded Object Counting
a VLM benchmark testing spatial reasoning by making the models count objects under occlusion
Atin Pothiraj, Elias Stengel-Eskin, Jaemin Cho, Mohit Bansal
Preprint Cite Code
CAPTURe: Evaluating Spatial Reasoning in Vision Language Models via Occluded Object Counting
Ctrl-Adapter: An Efficient and Versatile Framework for Adapting Diverse Controls to Any Diffusion Model
A plug-and-play framework that reuses any existing ControlNet for any video/image diffusion model
Han Lin*, Jaemin Cho*, Abhay Zala, Mohit Bansal
Preprint Cite Code Project
Ctrl-Adapter: An Efficient and Versatile Framework for Adapting Diverse Controls to Any Diffusion Model
DataEnvGym: Data Generation Agents in Teacher Environments with Student Feedback
A new testbed of teacher environments for data generation agents for diverse tasks.
Zaid Khan, Elias Stengel-Eskin, Jaemin Cho, Mohit Bansal
Preprint Cite Code Project
DataEnvGym: Data Generation Agents in Teacher Environments with Student Feedback
M3DocRAG: Multi-modal Retrieval is What You Need for Multi-page Multi-document Understanding
Multi-modal RAG framework and dataset for multi-page multi-document understanding
Jaemin Cho, Debanjan Mahata, Ozan İrsoy, Yujie He, Mohit Bansal
Preprint Cite Code Project
M3DocRAG: Multi-modal Retrieval is What You Need for Multi-page Multi-document Understanding
SELMA: Learning and Merging Skill-Specific Text-to-Image Experts with Auto-Generated Data
SELMA improves T2I models by fine-tuning on automatically generated multi-skill image-text datasets, with skill-specific LoRA expert learning & merging. - NeurIPS 2024
Jialu Li*, Jaemin Cho*, Yi-Lin Sung, Jaehong Yoon, Mohit Bansal
Preprint Cite Code Project
SELMA: Learning and Merging Skill-Specific Text-to-Image Experts with Auto-Generated Data