Support for new FLIR annotations format

[daddydrac]

Add in support to convert FLIR to COCO

[FranciscoReveriano]

?

[daddydrac]

@FranciscoReveriano apparently FLIR reformatted to one long file with the annotations nested in-between other things that arent needed.

[FranciscoReveriano]

That sounds silly. I don't know what FLIR is thinking. I have been busy with proprietary datasets and haven't had the time to look into FLIR very much. I am planning sometime this month to run it. But I hear alot of people having problems with it.

Great marketing for them. I will try it when I finally get a few hours to run your code.

[joel5638]

@FranciscoReveriano It helps a lot. waiting for your support in converting FLIR json to coco format

[FranciscoReveriano]

@joehoeller I will be publishing some code to read that long JSON file. Your Dark Chocolate is not designed to read from text files correct?

[daddydrac]

It takes individual coco json files and outputs Darknet annotation format.

I guess FLIR is one big JSON file now with other nested arrays that are not needed.

I’ve been so busy working I have not had time to look into it in-depth. However, if you use PyDash reduce, you can reduce and map FLIR down into multiple COCO annotations.

Then re-use the math and functions to make Darknet.

[FranciscoReveriano]

I am currently setting a pipeline to make that work. Well Convert from that huge JSON into txt files and then COCO normalized txt files. That's why I was asking.
I might try your format if I have time.