Add soccerplayer identification ability

[piercus]

Functionnal context

Hello Tania, i need to build database for urbansoccer player. I would like to have thousands of 'box' images of each player in a game and have them grouped by player

We are having thousands of images of each player, but we are not able to group them by player yet.

I would like you to help me to improve term-annotator so we can split all the boxes of a football match into 10 category (each per player) easily.

Actual solution

Basically, we will create 12 categories to start with :

• Player 1 - Team 1

• Player 2 - Team 1

• Player 3 - Team 1

• Player 4 - Team 1

• Player 5 - Team 1

• Player 1 - Team 2

• Player 2 - Team 2

• Player 3 - Team 2

• Player 4 - Team 2

• Player 5 - Team 2

One specific category :

• Referee

And 2 garbage categories

• Not a clean player (track is not right)
• not sure

In current urbansoccer state machine we have tracks, and each track is about 5 to 20 seconds long

I will need you to generate one video per track, with the images from both cameras on corresponding track.

For this, please create a lambda function (cause this might run on remote) called gifFromTrack that takes a s3 event of a track file creation event and build a gif for this track file into rushBucket/<client>/<user>/<rush_id>/trackGif/<track_id>.gif.

Please create an issue into sls-noe-rush-salad-analysis and create corresponding unit tests and code from urbansoccer latest branch.

Then you will try to use the current term-annotator and put those trajectories videos into one of the 12 classes.

Next step

Then we will discuss how to improve term-annotator to make it better.

[TaniaPash]

Status

branch: https://github.com/teamklap/term-annotator/tree/tan/9-ident-ability term-annotator was updated with the following features:

the main menu looks like this now.

once the player is annotated: png from 1st gif-frames is stored into annotated-folder and set as the thumbnail.

Steps to reproduce

0. cd term-annotator
1. git checkout tan/9-ident-ability
2. npm i
3. download gifs to annotate mkdir ../gifs-to-annotate && aws s3 cp s3://dev-bri-bucket-rush-eu-west-1/urbansoccer/20190922/154357-1/trackGif/ ../gifs-to-annotate --recursive
4. npm run start
5. annotated
6. check annotated files in ../annotated/****** folders.

Notes:

• configurable input folder, app's packaging is in PR: https://github.com/teamklap/term-annotator/pull/7
• the code is not mature. I was trying to implement these changes as fast as i can. i'll update code once it is decided to use this annotation process.

Performance test

• it is still hard to identify a player with a thumbnail also. As the first frame of a gif does not always show player with a number. I think it will work out with 2 annotation circles: annotate as usual, re-annotate categories notSure and notClean
• as an option, the annotator can do a quick (fast) watch of the annotated video segment before annotation (it is like a suggestion )