Saturday, 29 December 2018

Easy Bounding Picture Annotation Box with a simple mod for VGGannotated image


Easy Bounding Picture Annotation Box with a simple mod for VGG annotated image

You've got a great new educational computer setup. Many cores, or a lot of memory, or NVIDIA 1080Ti or two, probably a Titan V. Nice Linux is all setup, Anaconda Python, TensorFlow, Keras, installed. You have a bunch of new image data and a "real" problem to solve! What now?


If your newly created real-world imagery data is very likely, you need to make some object annotations to get your training data in a usable state. How are you going to do this? Maybe you

People randomly push the Amazon Turkmen to do this for you using one of the bundles of comments available there.

Get the slave labor, ie graduate students, to spend 16 hours a day working on it.

Do it yourself (hoping to help friends, colleagues or co-workers).

I do not have graduate students, and the data for one of my projects is private, so it can not be "sourced" or transmitted to any online service. But, I have willing colleagues (I hope). It's just a matter of using a good, easy-to-use illustration tool that I can provide to my colleagues at work.

It turns out that there are lots of annotations tools for the pictures, but hardly any of them meet my needs! Just search online for photo annotations or see the Wikipedia menu.

The annotation tool's requirements are,

easy to use

Fast (I can not spend people spending dozens of hours working on it)

It can not require any software installation with dependencies like Python or Java etc.

They must be portable from work to home and do not require a web server.

Must be free (due to lack of official budget)

Use,

I need to create simple surround box coordinates with labels

I just need to label a few categories of objects

Did you mention quickly and easily ...

It turned out that it was difficult to find any illustrative tool that could meet these requirements. I found one that was "almost big". This post is about "very large" trim to "awesome" (for my own use case).

VGG Image Annotator (VIA)


VIA is an open-source image inspector that is very useful. It is one of many interesting projects under the active development of the Oxford University Group for Visual Engineering (VGG).

Some of the great features of VIA are,

HTML, CSS, and JavaScript files completely (no external JavaScript libraries)

It exists in one html file. This file can be opened in the browser and use it from the line.

They have a variety of shapes in the area (I only need rectangles)

...

VIA is easy to use and intuitive. However, version 1.0.x is slow to use. You can easily scroll through a list of images and simply use the mouse to draw the bond boxes, however, you must enter the region labels by typing them into an interface such as the spreadsheet in a different part of the GUI.

Here's an example,

There is no MOD interface image

My simple edit will add easy-to-use buttons placed under the image. It is very fast to use!

Modified VIA image


We hope the screenshot above gives you an idea of ​​how fast and easy you can use it. You can place one hand on the mouse to draw the surrounding squares and click the label buttons, and on the other hand the right arrow key to change to the next image. In the next section, I'll show you a simple adjustment to the VIA source to add the label buttons.

With this simple adjustment, I discovered that I could work square boxes and posters with 3 rows at a rate of 15-20 pictures per minute.

VIA Buttons-for-Labels Modification


The amendment I describe here is VIA version 1.0.6, which was released on June 15, 2018.

VIA 1.0.6 is included in a single html file that does not load any external JavaScript libraries. Is approximately 5,000 lines of mostly JavaScript. It took some time to figure out how to add the jobs you wanted. What I did was not elegant or a good coding practice. It's just a "quick breakthrough" to make them do what I want.

There are two things you added,

Set the html buttons directly after the "div" for the image pane.

The javascript function called by the buttons. This adds the surround box coordinates and annotations to the annotations file and updates the state of the GUI.

Add buttons

I will use the code from the mod I used to picture a guy for example a bike. You can add more or less buttons and any labels you need.

No comments:

Post a Comment