Lesson 1. Bounding box detection in fastai-1.x

Part 2 & Alumni (2018)
4

@ilya_bohaslauchyk My apologies for responding to an old conversation. Your comment was super helpful. I’ve been stuck on this for a bit.

Could you help understand why 112 makes sense here and if there is a way to approach the single object detection more efficiently for batch predictions?

I have a fairly brute force approach at the moment using fastai v1.

Here I’m simply trying to visualize the prediction coordinates on the validation dataset.

for idx, pred in enumerate(preds):
    if idx > 20: break
    coords = (pred.data + 1)* 112
    coords = coords.tolist()
    
    category = ll.data.valid_ds.y[idx].labels[0]
    img = ll.data.valid_ds[idx][0]        

    bbox = ImageBBox.create(*img.size, [coords], labels=[0], classes=[category])    
    _display_with_bbox(img, bbox)

Thanks.