goorealv3 / README.md
markytools's picture
Update README.md
ac7a9a5
metadata
dataset_info:
  features:
    - name: image
      dtype: image
    - name: split
      dtype: string
    - name: width
      dtype: int64
    - name: height
      dtype: int64
    - name: bboxes
      dtype: string
    - name: labels
      dtype: string
    - name: gaze_item
      dtype: int64
    - name: gazeIdx
      dtype: int64
    - name: gaze_cx
      dtype: int64
    - name: gaze_cy
      dtype: int64
    - name: hx
      dtype: int64
    - name: hy
      dtype: int64
    - name: seg
      dtype: string
    - name: occluded
      dtype: bool
    - name: person_num
      dtype: int64
    - name: cam_num
      dtype: int64
  splits:
    - name: test
      num_bytes: 6289998121.391
      num_examples: 7391
  download_size: 6286282416
  dataset_size: 6289998121.391

The dataset features/columns here are almost similar to the original github instruction (please read the github documentation first to understand the dataset): https://github.com/upeee/GOO-GAZE2021/blob/main/dataset/gooreal-download.txt

To download gooreal in huggingface, run the code below (https://huggingface.co/docs/datasets/v1.10.0/loading_datasets.html#from-the-huggingface-hub):

from datasets import load_dataset
dataset = load_dataset("markytools/goorealv3")

The image datasets will be stored in ""~/.cache/huggingface", so you need to delete the files here if you want to free up space.
The "bboxes" and "labels" features are in string format, so you can use the code below to convert the string into list:
import ast
listOfBboxes = ast.literal_eval(dataset["test"]["bboxes"][0])

The feature "seg" is now in string format instead of numpy ndarray. This is an optional feature, and you can manually download the files here (https://huggingface.co/datasets/markytools/goosegmv3) using wget commandline. The files are in .npy so load it using np.load (https://numpy.org/doc/stable/reference/generated/numpy.load.html).