How to view data from https://sites.google.com/site/costardataset

view_costar_block_stacking_attempt.py

import os
import h5py
import io
import matplotlib.pyplot as plt
import numpy as np
from PIL import Image


def JpegToNumpy(jpeg):
    stream = io.BytesIO(jpeg)
    image = Image.open(stream)
    return np.asarray(image, dtype=np.uint8)


def ConvertImageListToNumpy(data):
    images = []
    for raw in data:
        img = JpegToNumpy(raw)
        images.append(img)
    return np.array(images, dtype=np.uint8)


# First download an example like the one linked here
# https://archive.org/download/johns_hopkins_costar_dataset/blocks_only/2018-05-23-18-00-08_example000001.success.h5f
# Then update the path below to reference your downloaded file
path = ('~/.keras/datasets/costar_block_stacking_dataset_v0.4/blocks_only/'
        '2018-05-23-18-00-08_example000001.success.h5f')

# Open the h5f file using h5py
with h5py.File(os.path.expanduser(path), 'r') as data:
    # You can use data[key] to access a particular feature column.
    # Refer to "Dataset Features and Time Steps" section for
    # more information on available features.
    # Here, we get the frames for each goal in the file
    goal_frames = np.unique(data['gripper_action_goal_idx'])

    # Get the images corresponding to the goals
    image_list = np.array(data['image'])[goal_frames]

    # The images are stored as JPEG files.
    # We now use functions defined above to convert the images to np arrays
    images = ConvertImageListToNumpy(image_list)

    # Turn the images into a tile
    tiled_images = np.squeeze(np.hstack(images))

    # Show the images with pyplot
    plt.imshow(tiled_images)
    plt.show()