Datasets:

Modalities:
Image
Text
Formats:
parquet
ArXiv:
Libraries:
Datasets
Dask
License:
jan-hq's picture
Add task category, paper link, and license (#2)
3be2f49 verified
metadata
dataset_info:
  features:
    - name: image
      dtype: image
    - name: conversations
      dtype: string
  splits:
    - name: train
      num_bytes: 14118375955.86
      num_examples: 99000
    - name: test
      num_bytes: 143139216.14
      num_examples: 1000
  download_size: 14073666948
  dataset_size: 14261515172
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
task_categories:
  - robotics
license: apache-2.0

This dataset is associated with the paper PoseLess: Depth-Free Vision-to-Joint Control via Direct Image Mapping with VLM.