arxivcs.CV2026-06-29
OmniCoT: A Benchmark for Global and Multi-Step Panoramic Reasoning
Haocong He, Chenfei Liao, Zichen Wen, Zihao Dongfang, Xu Zheng, Bin Ren, et al.
Multimodal Large Language Models (MLLMs) have demonstrated promising spatial reasoning capabilities, while these abilities remain underexplored in the emerging visual modality of panoramic imagery. The full 360°$\times$180° field of view of panoramas essentially supports complex…