Generate TFRecord file with TensorFlow.

This is a collection of handy Python scripts related with TensorFlow TFRecord file generation.

• split_data.ipynb: A notebook shows how to split the full dataset into train, validation and test subsets.
• generate_tfrecord.py: Generate a TFRecord file.
• view_record.py: View the contents of a TFRecord file.

TensorFlow

pip3 install tensorflow

• numpy
• pandas
• OpenCV (only if you need to run view_record.py to preview images).

Git clone this repo then you are good to go.

git clone https://github.com/yinguobing/tfrecord_utility.git

Assuming you have IBUG data organized in the following manner:

• /data/landmark/image Extracted face images.
• /data/landmark/mark Extracted facial landmarks in JSON files.

and you have list all the samples' name in a csv file:

/data/landmark/ibug.csv

and you want to put the generated TFRecord file here:

/data/landmark/ibug.record

Finally run the script like this:

python3 generate_tfrecord.py \
  --csv /data/landmark/ibug.csv \
  --image_dir /data/landmark/image/ \
  --mark_dir /data/landmark/mark/ \
  --output_file /data/landmark/ibug.record

The generated file ibug.record should be found.

This project is licensed under the MIT License - see the LICENSE.md file for details

• The official TensorFlow data tutorial.