tf-human-action-datasets

repository organization

• datasets/: all stuff related with datasets' loading and manipulating.
• datasets/common/: utilities related with joint and labels shared across datasets.
• datasets/montalbanov2/: functions and joint definitions from Montalbano V2 dataset.
• datasets/montalbanov2/compute_average_lengths.py: script used to compute average lengths of each skeleton's connection. This measurements are needed in normalization process.
• datasets/ntu_rbgd/: script to load and generate tfrecords to work with NTU-RBGD dataset.
• datasets/tfrecords/: set of utilities to encode and decode tfrecords data structures.
• datasets/tfrecords/: set of utilities to encode and decode tfrecords data structures.
• etc/conf/: examples of configurations used in some scripts.
• etc/experiments/gesture_localization/: folders containing configuration and results of experiments about gesture localization process. They're separated in 3 types: train_*, eval_* and search_*.
• examples/: scripts to validate implementations.
• models/base/: general stuff used to work with all models.
• models/gesture_localization/: scripts used to prepare data, train and validate models to address gesture localization problem.
• models/options/: contains a *.proto file with all configurations used by the scripts from this repository, as well as a helper function to load and validate configurations.
• models/skeleton_net/: scripts created when trying to build a model based on SkeletonNet using NTU-RGBD dataset.
• tf_patch/: a monkey patch used to fix a tensorflow issue, which is probably already fixed.
• utils/: scripts and classes used all over this repository.

references

tfrecord

• https://github.com/chiphuyen/stanford-tensorflow-tutorials/blob/master/2017/examples/09_tfrecord_example.py

• https://github.com/tensorflow/tensorflow/blob/v1.12.0/tensorflow/examples/how_tos/reading_data/fully_connected_reader.py

• http://warmspringwinds.github.io/tensorflow/tf-slim/2016/12/21/tfrecords-guide/

• https://medium.com/@dikatok/making-life-better-read-easier-with-tensorflow-dataset-api-fb91851e51f4