How to use extract_yt8m_features to to produce youtube-8m compatible .tfrecord files?

[allenday]

youtube-8m repo indicates that mediapipe feature extractor should be used instead of extract_tfrecords_main.py:
https://github.com/google/youtube-8m/tree/master/feature_extractor
Presumably this refers to extract_yt8m_features, described here:
https://github.com/google/mediapipe/tree/master/mediapipe/examples/desktop/youtube8m

The mediapipe feature extractor outputs a single .pb file per input video. Two questions:

1. how to output a .tfrecord file of the format required by youtube-8m train.py?
2. how to incorporate a CSV of labels, as used by extract_tfrecords_main.py?

Essentially, I am asking for equivalent steps from mediapipe to perform the youtube-8m command:

python extract_tfrecords_main.py --input_videos_csv /path/to/vid_dataset.csv \
    --output_tfrecords_file /path/to/output.tfrecord

[allenday]

youtube-8m issue created and linked here:
google/youtube-8m#124