sermanet/multiview-pouring

MultiView Pouring Dataset, v1.0.

by Pierre Sermanet, Corey Lynch, Jasmine Hsu and Eric Jang

License

This data is licensed by Google Inc. under a Creative Commons Attribution 4.0 International License.

Downloading

Because of some downloading issues for a specific file, the file was split in two, call https://huggingface.co/datasets/sermanet/multiview-pouring/blob/main/tfrecords/test/whiteorange_to_clear1_real_combining.sh to recombine the parts.

Structure

Training, validation and test sets are provided with the following structure:

videos/: original .mov videos (1080x1920 resolution, non-rotated).

tfrecords/: TensorFlow-ready files, where each video is a SequenceExample object (https://www.tensorflow.org/api_docs/python/tf/train/SequenceExample) containing all frames of the videos, compressed as a jpeg-encoded string. Each original frame is rotated upright and downsampled to 360x640 pixels.

labels/: For each tfrecord video in tfrecords/val and tfrecords/test, we provide a tfrecord with the same filename that contains the labels for each frame of that video.

Generating tfrecords.

Although tfrecord files are already available here, one can regenerate them from the original videos using the following command:

set=multiview-pouring && videos=~/data/tcn/ && tcn/videos_to_tfrecords --logtostderr --output_dir ~/data/tcn/tfrecords/$set --input_dir $videos/datasets/$set/ --rotate 90 --resize_min_edge 360