Skip to content

Rhoban/tagger

master
Switch branches/tags
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
 
 
bin
 
 
 
 
 
 
 
 
src
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Rhoban patches tagger

Introduction

The tagger is a tool allowing you to upload several patches (small images) and tag them under some categories in order to train neural networks

imgs/tagger.png

Installing

Cloning and getting dependencies

The current version is based on the Symfony framework and require a MySQL database.

First, clone this repository and run the composer installation:

git clone https://github.com/Rhoban/tagger.git
cd tagger
composer install

Note: composer can be obtained on getcomposer.org or via apt install composer on a debian installation

Configuring the database

You then need to configure the app, edit the .env file and change the DATABASE_URL so that it matches your database configuration.

You can now create the database schema using:

./bin/console doctrine:schema:create

Getting the first administrator

You can now deploy and run the application (reaching the Symfony public/ directory from your web server supporting PHP), and register your first account.

To promote the first admin, you can use the Friends Of Symfony command line:

./bin/console fos:user:promote

And then enter the username, and ROLE_ADMIN like this:

Please choose a username: gregwar
Please choose a role: ROLE_ADMIN

You might need to logout and login again to have this taking effect.

Using

Managing categories

First, you need to create categories, log in on any admin account and click on the category menu on the top.

The categories are basically things that you want to tag. In our robots at the RoboCup we tag for instance balls and goal posts.

Sessions and sequences

Next, you need to upload log sessions that you want to tag. A session can be any way of separating your batches of patches. A session contain several sequences that are typically shot sequentially, and can be used to know where each patch come from more accurately.

For example:

  • Sesion 1: "Training field at our lab"
    • Sequence 1: "Robot 1 at 10:01"
    • Sequence 2: "Robot 1 at 10:09"
    • Sequence 3: "Robot 2 at 10:12"
  • Session 2: "Field A at RoboCup 2017"
    • Sequence 1: "Robot 1 at 12:31"
    • Sequence 2: "Robot 2 at 12:45"

Thus, we can for instance disable the session that was using for training at our lab when we want only to tag the patches at the competition. Moreover, we can also remove sequences from the Robot 1 if we note that there was a big mistake in the patch extraction.

To create sequence, you need to upload patches using the upload process

Uploading

The archive can have any structure, as long as it ends with directories with categories names, like:

robot1/
    17h59/
        goal/
            1.png
            2.png
            3.png
        ball/
            1.png
            2.png
            3.png
    18h05
        goal/
            1.png
            2.png
            3.png

Will result in having two sequences, one named robot1_17h59 containing patches for goal< category and patches for ball category, and one named >robot1_18h05 containing patches for goal category.

Note: if you want to have the progress bar to work, you have to enable the session.upload_progress.enabled in the PHP configuration

Tagging

When tagging, a grid of images is presented to you:

imgs/grid.png

Images you don't click will be registered as negative for the current category. With one click, you mark the images as positive for the class. With two click, you mark it as unknown which means that you don't want it to be used for training because it's ambiguous (for instance, if you have images that will not appear in real situations):

imgs/grid_tagged.png

You can then click OK and tag the next grid.

Note: you can choose the size of the images and the grid in "My settings", this can be handful to adapt the size of the grid to your screen or the device you are currently using

Consensus

To have images actually accepted in one category, you need it to have consensus, which mean that:

  • At least CONSENSUS_MIN_USERS tagged it, default is 2
  • There is one class (yes, no or unknown) accepted by a ratio higher than CONSENSUS_THRESHOLD, default is 0.6 (60%)

You can change these values by putting it in your .env file, for example:

CONSENSUS_MIN_USERS=1
CONSENSUS_THRESHOLD=0.6

Will make that one user alone will be able to trigger a consensus

Note: if you update the values of the consensus settings, you might also want to run ./bin/console app:consensus to update the statistics

Training

This feature is optional, add TRAINING=1 in the .env file to enable it.

When fresh users registers, they are marked as not trained, meaning that they are not allowed to tag unless they completed a training session.

To achieve this:

  • Tick the "training" box when editing sessions that you want to include in the process of training new users
  • Tag the images from these sessions
  • You can, if you want, disable these sessions so they are not in the standard tag process (and let the training flag)

Notes: admin doesn't need to be trained, it is also possible to manually mark someone as trained in the administration of users.

LICENSE

This is under MIT license. Please refer to the LICENSE file for more information

About

Collaborative patches tagger

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published