C4 Week 3 assignment Auto Detection

Course Q&A Deep Learning Specialization Convolutional Neural Networks
1

Yolo non max suppression
I am getting that yolo_filter_boxes is correct, function iou is correct but yolo_non_max_suppression is throwing the following error:

InvalidArgumentError: Value for attr ‘T_threshold’ of int32 is not in the list of allowed values: half, float
; NodeDef: {{node NonMaxSuppressionV3}}; Op<name=NonMaxSuppressionV3; signature=boxes:T, scores:T, max_output_size:int32, iou_threshold:T_threshold, score_threshold:T_threshold → selected_indices:int32; attr=T:type,default=DT_FLOAT,allowed=[DT_HALF, DT_FLOAT]; attr=T_threshold:type,default=DT_FLOAT,allowed=[DT_HALF, DT_FLOAT]> [Op:NonMaxSuppressionV3]

that the iou_threshold dtype is incorrect.

Any ideas what could be wrong?

[Coursera | Online Courses & Credentials From Top Educators. Join for Free | Coursera]

2

Windows 10 python 3 in jupyter notebook

3

Two questions:

  • Are you working in Coursera Labs, or on some other platform?
  • Are you using the max_boxes_tensor variable when you call non_max_suppression()?
4

Yes working in Coursera Labs, I use max_boxes_tensor variable when I call non_max_suppression but have also tried it with max_boxes and with max_output_size=max_boxes

5

Only one of those choices is correct. It’s the first one.

6

image
image733×521 26.6 KB

… where the max_output_size is provided by the max_boxes_tensor.

7

Yes but it still generates the same error message. I don’t ser the iou_threshold dtype so i am confused by the error message.

8

see

9

And I have tried all three and still get the same error message

10

Please click on my ID and use the “Message” button, to send me a screen capture image of this entire function.

11

are you using max_boxes_tensor instead of boxes in the tf.image.non_max_suppression function call?

12

Thanks for your suggestion as to the nature of the bug here, but please note that this thread is more than 6 months old. It’s pretty likely that the author of the original thread has found a solution. Of course your comment could still benefit someone new who finds this thread through the forum search.