/ILSVRC2015VID_TFRecords

A python library for generating TFRecords from ILSVRC2015 VID dataset

Primary LanguagePython

ILSVRC2015VID_TFRecords

This repository is used to generate sequencial TFRecords files from raw ILSVRC2015VID datasets. The generated record files can be used to train networks used for Visual Tracking, Object Detection Tasks, e.t.c.

Requirements

  • python3.5 or later
  • tensorflow-gpu == 1.4.1 (or tensorflow == 1.4.1)
  • tqdm
  • ILSVRC2015 VID dataset

run pip install -r requirements.txt to install required packages.

go to ILSVRC2015 for more information and download the dataset.

Create fixed length TFSequenceExample

run scripts/main_create_fixed_len_examples.py

use tfhelper.input_fn.fixed_len_input_fn for tf.Estimator API