FLAT-PACK-BENCH: The Benchmark Dataset

The full text of questions and images that have been used for prompts are provided in this dataset, structured along with useful metadata.

The directory structure is <question-type>/data/<furniture-category>/<furniture-name>/<video-id>/

Within each such directory is a questions.jsonl file that contains on every line a separate question that refers by file name to images that are provided in prompt_images.

The videos are provided by the base dataset using the same <video-id>. As described in their project repository, the videos can be obtained at https://purl.stanford.edu/sg200ps4374 . Please see the CREDITS.md file for a full citation.

manual_seg_masks contains our manually annotated segmentation masks.