Datasets:

MM-Hallu
/

ROPE

dataset_info:
  features:
    - name: image
      dtype: image
    - name: label
      dtype: string
    - name: split_type
      dtype: string
    - name: objects
      dtype: string
    - name: object_set
      dtype: string
    - name: relations
      dtype: string
    - name: size
      dtype: string
  configs:
    - config_name: default
      data_files:
        - split: train
          path: train-*.parquet
        - split: validation
          path: validation-*.parquet

ROPE

ROPE is an object detection benchmark combining images from ADE20K and COCO with bounding box annotations. 5,161 examples across 5 spatial relationship patterns (AAAAB, BAAAA, homogenous, heterogenous, mixed).

Fields

Field	Description
image	Input image (with bounding box overlays)
label	Source dataset: `ADE` or `COCO`
split_type	Relationship pattern
objects	JSON-encoded list of objects with bounding boxes
object_set	JSON-encoded list of object names
relations	JSON-encoded spatial relations
size	JSON-encoded image dimensions

Converted from sled-umich/ROPE.