Add a COCO to YOLO converter.

[Abdul-Mukit]

Feature Description

A coco2yolo.py script that converts a COCO formatted detection or instance segmentation dataset into a yolo dataset.
Expected usage:
python coco2yolo.py --cocoroot <root path of coco dataset> --exportpath <root path of expected yolo dataset>
cocoroot will be the root path of coco dataset. This directory will contain "annotations" and "images" folders. The exportpath will be path where the converted data will placed. After conversion finishes the exportpath directory will contain two folders. labels and images. If the exportpath is same as cocoroot the folder images are not over-written.

I am working on a converter and will be making a PR soon. @henrytsui000 will appreciate your help in getting it evaluated and merged. Thanks.

Future Considerations

Currently, it seems that we are making a double effort in data_loader.py to support both types of datasets; COCO and YOLO. This is causing our effort to be diverted to fixing bugs like #67 #36.
Let us make this this converter script and reevaluate, if we should only support yolo formatted dataset. This should help us focus our efforts better.

[henrytsui000]

Hi,

Due to multiple deadlines, I've been quite busy these past few weeks. I will review it during my break time as soon as possible. I deeply appreciate your contribution and your willingness to help with this project!

Best regards,
Henry Tsui

[yjmm10]

In the past week, I have completed some work, to share:
The data set supports YOLO and COCO formats, and the configuration can specify multiple folders like YOLO, and supports mixing YOLO and COCO formats, so in #72 I might close MR And mention a new one.
I should add that for the image_id problem #67 mentioned earlier, I used a mapping relationship and returned it, which is not supported for YOLO. By default, the cache file is correct when it is generated for the first time and is not verified.
I will PR soon.

[Erotemic]

Note I have PR where there is some support for this:

#150