Detection of Pikachu on Android using Tensorflow Object Detection API
Branch: master
Clone or download
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
android updated android threshold and training number of epochs May 12, 2018
training updated android threshold and training number of epochs May 12, 2018
LICENSE Create LICENSE Jan 21, 2018
README.md Specified TensorFlow version Aug 23, 2018
detection_video.py added detection video script May 9, 2018
generate_tfrecord.py
pikachu_object_detection_evaluation.ipynb added training set and jupyter notebook Jan 21, 2018
xml_to_csv.py added script to generate dataset Jan 21, 2018

README.md

Detecting Pikachu on Android using Tensorflow Object Detection

Overview

This repo contains the code used in my experiment titled "Detecting Pikachu on Android using Tensorflow Object Detection". In this experiment, which is available here, I explained the many steps needed to train a custom object detection model using TensorFlow Object Detection API and how to deploy it in an Android device.

The project was done on TensorFlow 1.4

This project has been updated with a video detection feature. For a detailed explanation of why it was updated, and how the video detection was done, check out my second article titled Detecting Pikachu in videos using Tensorflow Object Detection

The code

The content of this repo is mostly divided in 4 parts

  • The directory android contains the 'gradle.build' file used to build the example TensorFlow provides, and the file 'DetectorActivity.java' which is the responsible of performing the detection in the app
  • The directory training has the final models my training produced, as well as the pipeline configuration file required for the training.
  • The script detection_video.py which is used for performing the detections in videos
  • The rest of the files are those scripts needed to prepare the dataset.

Instructions

All this code by itself does not do anything. It must be used in combination with the Object Detection API. The report linked above has all the instructions on how to use the code alongside the API.