A dataset generation framework for spatial reasoning tasks involving camera viewpoint prediction and ordering around static 3D objects. This project generates multimodal chain-of-thought reasoning traces that teach models how camera views change during orbital rotation.
Task 1: Camera View Prediction - Given an initial view and rotation parameters (angle +… See the full description on the dataset page:
https://huggingface.co/datasets/Caesarrr/interleaved-umm-new.
