ObjDetection: make the prediction intuative

[Borda]

🚀 Feature

refactor the detection predictions

Motivation

from example:

predictions = model.predict(["data/coco128/images/train2017/000000000625.jpg"])
print(predictions[0])

now you get a single image of Coco128

[[{'boxes': [tensor(260.3752), tensor(122.2690), tensor(399.8370), tensor(428.3777)], 'scores': tensor(0.0854), 'labels': tensor(1)}, {'boxes': [tensor(167.9021), tensor(182.6161), tensor(398.0537), tensor(413.8837)], 'scores': tensor(0.0761), 'labels': tensor(1)}, {'boxes': [tensor(331.9625), tensor(134.9533), tensor(468.3016), tensor(446.)], 'scores': tensor(0.0662), 'labels': tensor(1)}, {'boxes': [tensor(405.6042), tensor(144.5019), tensor(539.5533), tensor(446.)], 'scores': tensor(0.0651), 'labels': tensor(1)}, {'boxes': [tensor(295.9349), tensor(162.3046), tensor(530.8933), tensor(385.1946)], 'scores': tensor(0.0629), 'labels': tensor(1)}, {'boxes': [tensor(232.7350), tensor(168.4053), tensor(355.7172), tensor(438.0991)], 'scores': tensor(0.0615), 'labels': tensor(1)}, {'boxes': [tensor(289.9357), tensor(155.3581), tensor(360.9719), tensor(322.8009)], 'scores': tensor(0.0549), 'labels': tensor(1)}, {'boxes': [tensor(155.4224), tensor(150.9095), tensor(294.6923), tensor(446.)], 'scores': tensor(0.0535), 'labels': tensor(1)}, {'boxes': [tensor(283.4752), tensor(176.3774), tensor(421.8001), tensor(363.5930)], 'scores': tensor(0.0509), 'labels': tensor(1)}], [], []]

my confusion is:

• key boxes but it is just a single box
• box, why the list of tensors with a single number, rather one tensor of 4 elements or just numbers
• what are the very last empty lists?
• what is the meaning of the box numbers? x, y, w, h or x_c, y_c, w, h?

Pitch

More intuitive prediction understanding

Alternatives

implement also prediction visualisation

Additional context