Error at # Use tf.gather() to select only nms_indices from scores, boxes and classes

Course Q&A Deep Learning Specialization Convolutional Neural Networks
1

Hi,

I get the following error when using tf.gather

InvalidArgumentError: Value for attr 'Tindices' of float is not in the list of allowed values: int32, int64

I see that in a post from February, it’s suggested to use max_boxes_tensor instead of max_boxes, when calling tf.image.non_max_suppression, but with the latest implementation of this lab, there is no max_boxes_tensor, so I wonder what am I missing. Any suggestions are appreciated!

Thank you,
Leszek

2

That’s obsolete information.

When you call tf.gather(), use the selected indices and the nms indices labels.
image

3

Perhaps the issue is the code you’re using for nms_indices_label = ... or selected_indices = ...

4

so
1/ for nms_indices_labels I use tf.image.non_max_suppression
2/ to nms_indices I append result of tf.gather with ‘selected_indices’ (as indices) and nms_indices_label (as params)

Any obvious problems there? :slight_smile:

5

1/ OK given you provide the appropriate four parameters.
2/ OK.

6

1/ well, there are the following 4 parameters :slight_smile: - boxes, scores, max_boxes and iou_threshold :slight_smile:

interesting thing, tf.gather works for me with nms_indices and classes. Not with boxes or scores :confused:

8

Nope, those should be boxes_label and scores_label.

2 Likes
9

ha, this helped. Interesting thing, I had to switch selected_indices
and nms_indices_label in tf.gather. There is clearly something I dont understand about tf.gather :smiley:

thank you for your help and happy 2025! :slight_smile: that’s enough for 2024 for me, going to jump on the rest of the assignment tomorrow!

1 Like