W3 A1 | I wonder about tf.image.non_max_suppression()

In Week3 first programming assignment, we selected the max_boxes = 10. If number of detected boxes is greater than max_boxes after all calculations of nms, how to choose algorithm the rest boxes to remove? I tried to find out that in tf docments and google, but I coudn’t. Please help me.

Hello, Rlaskan.

This assignment belongs to Course 4, week 3 A1. I have changed that for you.

You have to check what criteria are you following. Here, we are selecting the number of detected boxes based on:

I think, it should not matter how many boxes are being detected until they pass those two criteria.

Well, other mentors can also pitch in the related thoughts here. It would be great to know their insights as well.

First, take a look at the description of the function in the TF doc

Greedily selects a subset of bounding boxes in descending order of score.

Then, the impact of max_output_size is provided in the description of the function return…

A 1-D integer Tensor of shape [M] representing the selected indices from the boxes tensor, where M<= max_output_size

My emphasis added

It’s just truncation

1 Like