[Detection Dataset Support] Add direct support for S3 detection dataset items in `draw_segmentation_mask` and `draw_bounding_boxes`

[cregouby]

Current situation

The coco dataset example seem to highlight how unaligned is the dataset item and the function to visualize it :
It requires

• image conditionning
• label conditionning
• boxes conditionning
• parsing of each of the 3 to the drawing function

suggestion

in order to make the COCO dataset more accessible, we could also manage the coco dataset item as a S3 class that would be directly managed by draw_bounding_boxes() and draw_segmentation_mask() . This would turn the example into the very straigthforward

coco <- coco_detection_dataset( root = "~/data", train = FALSE, year = "2017", download = TRUE)
item <- coco[1]
segmented_item <- draw_segmentation_mask(item)
tensor_image_browse(segmented_item)

@koshtiakanksha, what is your view on this ?

[koshtiakanksha]

@cregouby Thanks for the suggestion! I agree that this would make it much easier to use the dataset and run the examples. Having a custom class for each sample sounds like a good idea, and it would help the draw_ functions work more smoothly. I will start looking into how to add this and update the drawing functions to support it.

Let me know if you have any specific guidelines to get started.

[cregouby]

I would recommand the S3 chapter in advanced R

[cregouby]

fixed by #175