DeepPanel for Android

Android library used to implement comic vignettes segmentation using a machine learning method named deep learning.

License

License

GroupId

GroupId

com.github.pedrovgs
ArtifactId

ArtifactId

deeppanel
Last Version

Last Version

0.0.1
Release Date

Release Date

Type

Type

aar
Description

Description

DeepPanel for Android
Android library used to implement comic vignettes segmentation using a machine learning method named deep learning.
Project URL

Project URL

https://github.com/pedrovgs/DeepPanelAndroid
Source Code Management

Source Code Management

https://github.com/pedrovgs/DeepPanelAndroid

Download deeppanel

How to add to project

<!-- https://jarcasting.com/artifacts/com.github.pedrovgs/deeppanel/ -->
<dependency>
    <groupId>com.github.pedrovgs</groupId>
    <artifactId>deeppanel</artifactId>
    <version>0.0.1</version>
    <type>aar</type>
</dependency>
// https://jarcasting.com/artifacts/com.github.pedrovgs/deeppanel/
implementation 'com.github.pedrovgs:deeppanel:0.0.1'
// https://jarcasting.com/artifacts/com.github.pedrovgs/deeppanel/
implementation ("com.github.pedrovgs:deeppanel:0.0.1")
'com.github.pedrovgs:deeppanel:aar:0.0.1'
<dependency org="com.github.pedrovgs" name="deeppanel" rev="0.0.1">
  <artifact name="deeppanel" type="aar" />
</dependency>
@Grapes(
@Grab(group='com.github.pedrovgs', module='deeppanel', version='0.0.1')
)
libraryDependencies += "com.github.pedrovgs" % "deeppanel" % "0.0.1"
[com.github.pedrovgs/deeppanel "0.0.1"]

Dependencies

compile (4)

Group / Artifact Type Version
org.jetbrains.kotlin : kotlin-android-extensions-runtime jar 1.3.72
org.jetbrains.kotlin : kotlin-stdlib jar 1.3.72
androidx.core » core-ktx jar 1.3.1
org.tensorflow : tensorflow-lite jar 2.2.0

Project Modules

There are no modules declared in this project.

DeepPanelAndroid Build, lint, and test

Android library used to implement comic vignettes segmentation using a machine learning method named deep learning.

DeepPanel let's you extract all the panels' location from a comic page using a machine learning model trained here. This Andorid library does not use Open CV at all, this means the size of the final app you generates will be as small as possible. We've optimized our model in terms of size and performance and implemented the computer vision algorithms we needed without any third party library to optimize the final library size and performance. Powered by TensorFlow lite, our already trained panels' segmentation model and some native code, DeepPanel is able to find all the panels' location in less than a second.

Keep in mind, this artificial intelligence solution is designed to simulate the human reading behavior and it will group panels that are related as a human would do while reading a comic book page. When analyzing a page we do not use image processing algorithms in order to find the panels. Our solution is based on deep learning. We do not use contour detection algorithms at all even when the samples analysis can look similar. If you are looking for a panels contouring solution you can use Open CV or any other image analysis tools like this project does.

Original Analyzed
page page
page page

In case you are looking for the iOS version of this project, you can find it here.

All the pages used in our sample project are created by Peper & Carrot which is a free comic book you should read right now ❤️

Usage

Dependency

Include the library in your build.gradle

dependencies{
    implementation 'com.github.pedrovgs:deeppanel:0.0.1'
}

To start using the library you just need to call DeepPanel.initialize with a valid Context. You can initialize this library from any Activity or from your Application class:

class MyApplication : Application() {
    override fun onCreate() {
        ....
        DeepPanel.initialize(this)
        ....
    }
}

Once you've initialized the library get any comic book page, transform it into a Bitmap instance and extract the panels' information using extractPanelsInfo method as follows:

val deepPanel = DeepPanel()
val result = deepPanel.extractPanelsInfo(bitmapSamplePage)
result.panels.panelsInfo.forEach { panel ->
            Log.d("DeepPanel", """Left: ${panel.left}, Top: ${panel.top}
                    |Right: ${panel.right}, Bottom: ${panel.bottom}
                    |Width: ${panel.width}, Height: ${panel.height}
                """.trimMargin())
        }

PredictionResult contains a list of panels with the position of every panel on the page and also a 2d matrix of integers with the following labels inside:

  • 0 - Label associated to the page of the content.
  • N - Content related to the same panel on the page.

Do not invoke DeepPanel extractPanelsInfo from the main app thread. Even when we can analyze a page in less than 400ms using a Pixel 4 as a reference device, you should not block your app UI at all. Our recommendation is to extract the analysis computation out of the UI thread using any threading mechanism.

Keep in mind for the first version of this library we consider every panel as a rectangle even if the content panel inside has a different shape. We will improve this in the future, for our very first release we've decided to use this approach because we will always represent the information of the panel on a rectangular screen.

Do you want to contribute?

Feel free to contribute, we will be glad to improve this project with your help. Keep in mind that your PRs must be validated by GitHub actions before being reviewed by any core contributor.

Acknowledgement

Special thanks to Asun, who never doubt we could publish this project ❤️

Developed By

Follow me on Twitter Add me to Linkedin

License

Copyright 2020 Pedro Vicente Gómez Sánchez

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

   http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

Versions

Version
0.0.1