cmusatyalab / tfrecord-scripts

Scripts for tfrecord files

• merge_tfrecords.py will combine the data from multiple tfrecord files into one file.
• verify_tfrecord.py will draw bounding boxes on images one at a time. Looking at a few of these images is a good sanity check before you start training.
• checkpoint_server.py will start a Gabriel server to process frames using an object detector from checkpoint files.
• saved_model_server.py will start a Gabriel server to process frames using an object detector from saved model files.
• remove_dup.py creates a new version of a tfrecord file where each image has a unique perceptual hash value.

Usage

Build the Docker container for TensorFlow Object detection by following these instructions.

Start the container with a volume map to the directory with your checkpoint files. Convert a checkpoint to a saved model by running:

python object_detection/exporter_main_v2.py --input_type image_tensor --pipeline_config_path <PATH TO pipeline.config> \
--trained_checkpoint_dir <Path to model_dir> --output_directory <Where to save model>

from ~/models/research in the container.

The two server scripts in this repository work with the clients in this directory.

Run python3 -m pip install imagehash to use remove_dup.py.