LocalJoost
commented
10 months ago YOLO works a bit odd. It does not draw a box around the recognized object - it has a lot of predefined boxes and tries to fit it a model it should recognize. In the case of my model, 1680 boxes
However, there might be boxes overlapping (only partially) and although I try to weed out overlapping boxes, there is a limit to when I decide it to be overlapping or not. It is possible there are multiple boxes around the airplane. I then try to find the one with the highest confidence. This has as drawback that two planes close to each other might be recognized as one. You can maybe fiddle around with the code to make this more optimal. I might do that one day.
The number is not the class, but just the box entry in the list of boxes found. At least, I hope so ;)
Hi, thank you so much for this amazing project and your explanations. I've tried it all, also on the HoloLens and it really works. However, I was wondering why, sometimes I get some unexpected results. To be precise, I obtained some console messages like these: "Found: 1 aircraft : 0,6699069" "Found: 2 aircraft : 0,7342909" "Found: 3 aircraft : 0,6712892" If the class is always the same, which is 0 as you explained, why do I also get these indexes? Thank you if you would like to write back!