Yes, that’s a good point. Sorry that I did not notice this point in Rishi’s reply. So the “meta” point is that you can just take as gospel truth everything you read here on the forums. The mentors try to keep things on the level, but we can’t catch everything. 
Also note that in the template code that was given to you for yolo_non_max_suppression, the logical structure of the code is correct including which parts of the logic are within the “for” loop and which are not.
I’m still very confused after reading and incorporating some of the logic here.
Error message for the last part:
This last case, there should be only one value for ‘scores’, right? But I kept getting this error message. I’ve tried multiple ways to debug it…
Other than this last case, my other test scenarios passed.
Please help.
You can see from the failing assertion that it is expecting two score values in that case. So what is different about this test case than the previous three? In this case, unlike the previous one, both classes are the same but you have a very high IOU threshold. If you get the same answer as the previous case, are you sure you are not hard-coding the IOU threshold value?
I have “iou_threshold=iou_threshold)” in the last part of “tf.image.non_max_suppression”…
it seems that there was another error when iou threshold at 0.9. Let me check. I’m also not quite sure what max_output_size should be… ‘1’ in this case?
The value of max_output_size is passed in as an argument to the top level function. Just use the passed value.
The way you are handling iou_threshold sounds correct.
