Course 4, Week 3, Part 1: yolo_filter_boxes function problem

I have a problem with the first function in Week3 assignment. I always get an error and I have no idea what I did wrong. The error is copied below. Can anyone help please, thanks!

TypeError Traceback (most recent call last)
in
3 boxes = tf.random.normal([19, 19, 5, 4], mean=1, stddev=4, seed = 1)
4 box_class_probs = tf.random.normal([19, 19, 5, 80], mean=1, stddev=4, seed = 1)
----> 5 scores, boxes, classes = yolo_filter_boxes(boxes, box_confidence, box_class_probs, threshold = 0.5)
6 print("scores[2] = " + str(scores[2].numpy()))
7 print("boxes[2] = " + str(boxes[2].numpy()))

in yolo_filter_boxes(boxes, box_confidence, box_class_probs, threshold)
33 # Step 3: Create a filtering mask based on “box_class_scores” by using “threshold”. The mask should have the
34 # same dimension as box_class_scores, and be True for the boxes you want to keep (with probability >= threshold) (≈ 1 line)
—> 35 filtering_mask = tf.math.greater_equal(box_class_scores,threshold)
36
37 # Step 4: Apply the mask to box_class_scores, boxes and box_classes (≈ 3 lines)

/opt/conda/lib/python3.7/site-packages/tensorflow/python/ops/gen_math_ops.py in greater_equal(x, y, name)
4053 _result = pywrap_tfe.TFE_Py_FastPathExecute(
4054 _ctx._context_handle, tld.device_name, “GreaterEqual”, name,
→ 4055 tld.op_callbacks, x, y)
4056 return _result
4057 except _core._NotOkStatusException as e:

TypeError: Cannot convert 0.5 to EagerTensor of dtype int64

The problem is in the tf.math.greater_equal call you are doing to compare a tensor elementwise with a scalar value. It is trying to do the analog of numpy “broadcasting” to convert the second argument into a tensor of the same type and shape as the first argument. But for some reason, it looks like your box_class_scores is of type int64, which causes it to throw that error. How could that happen? You can check it like this:

print(f"box_class_scores.shape {box_class_scores.shape}")
print(f"box_class_scores.dtype {box_class_scores.dtype}")

When I added those lines to my routine and ran the same test cell you are running, here’s what I got:

box_class_scores.shape (19, 19, 5)
box_class_scores.dtype <dtype: 'float32'>

What do you see when you try that?

1 Like

I guess this line should fix the problem → filtering_mask = (box_class_scores >= threshold)

I got the following lines:

box_class_scores.shape (19, 19)
box_class_scores.dtype <dtype: ‘int64’>

I guess I have the dimensions wrong.

1 Like

I solved the problem. Thanks!

It’s great to hear that you found the solution under your own power. I’m actually kind of curious how you got the box class scores to have an integer value. I would have thought they would naturally end up as floating point.

Update: Later in this thread, we see an example of how to cause that: by using the wrong input tensor in the reduce_sum to compute box_class_scores.

Hey, I am having the same shape issues as you had: box_class_scores.shape (19, 19), how did you solve this problem? (I did use axis =-1, but the recommended functions keeps shrinking the dimensions)

Hi, @XiaoXiaLiang!

Well, where does box_class_scores get generated? It depends on the value of box_scores, right? So maybe box_scores is the wrong shape. Debugging is always “one step at a time”. I added some more print statements like the ones that @gasperpavlic and I were talking about earlier and here’s what I get when I run the test cell for yolo_filter_boxes:

box_scores.shape (19, 19, 5, 80)
box_scores.dtype <dtype: 'float32'>
box_class_scores.shape (19, 19, 5)
box_class_scores.dtype <dtype: 'float32'>
scores[2] = 9.270486
boxes[2] = [ 4.6399336  3.2303846  4.431282  -2.202031 ]
classes[2] = 8
scores.shape = (1789,)
boxes.shape = (1789, 4)
classes.shape = (1789,)
 All tests passed!

What shape do you see for box_scores?

Hello, I like to thank you for taking the time to help me. It is very kind of you.

Shapes and data types:

box_scores.shape (19, 19, 5, 80)
box_scores.dtype <dtype: ‘float32’>

box_classes.shape (19, 19, 5)
box_classes.dtype <dtype: 'int64'>

box_class_scores.shape (19, 19)
box_class_scores.dtype <dtype: ‘int64’>

We can see here box_classes have the wrong shape and the data type is int64. This is because the tf.math.argmax() has saved the max value index in each of the 5 filters instead of the max value probability. I am unfamiliar with Keras and I am not sure how to solve this without adding a few lines of code to this function.

Maybe I have used the functions wrong!?

{moderator edit - solution code removed}

I used the wrong tensor in tf.math.argmax(). Thank you!

I think you mean that you used the wrong tensor in the reduce_max call, but I’m glad to hear that you found the solution.

4 Likes

You have the roles of arg_max and reduce_max reversed in computing box_classes and box_class_scores. arg_max is used to pick the classes, right?

Yes I reversed them because previously I was getting this output which is still 4-dimensional instead of 3-dimensional.

box_classes = tf.math.argmax(box_scores, axis=-1)
box_class_scores = tf.math.reduce_max(box_scores, axis=-1, keepdims=True)
print(f"box_class_scores.shape {box_class_scores.shape}")
print(f"box_class_scores.dtype {box_class_scores.dtype}")

box_scores.shape (19, 19, 5, 80)
box_scores.dtype <dtype: ‘float32’>
box_class_scores.shape (19, 19, 5, 1)
box_class_scores.dtype <dtype: ‘float32’>

Have you considered that maybe the problem is using keepdims in the way you did? Try it without that and see what happens.

I also have prints to show the shapes in my code. Here’s what I get when I run the test block for that function:

box_scores.shape (19, 19, 5, 80)
box_scores.dtype <dtype: 'float32'>
box_classes.shape (19, 19, 5)
box_classes.dtype <dtype: 'int64'>
box_class_scores.shape (19, 19, 5)
box_class_scores.dtype <dtype: 'float32'>
scores[2] = 9.270486
boxes[2] = [ 4.6399336  3.2303846  4.431282  -2.202031 ]
classes[2] = 8
scores.shape = (1789,)
boxes.shape = (1789, 4)
classes.shape = (1789,)
 All tests passed!
1 Like

Yes it is working for me now, thank you so much.

Sir,what is the correct one?

Hello everyone i still get this error can @paulinpaloalto please help me?

TypeError Traceback (most recent call last)
in
4 boxes = tf.random.normal([19, 19, 5, 4], mean=1, stddev=4, seed = 1)
5 box_class_probs = tf.random.normal([19, 19, 5, 80], mean=1, stddev=4, seed = 1)
----> 6 scores, boxes, classes = yolo_filter_boxes(boxes, box_confidence, box_class_probs, threshold = 0.5)
7 print("scores[2] = " + str(scores[2].numpy()))
8 print("boxes[2] = " + str(boxes[2].numpy()))

in yolo_filter_boxes(boxes, box_confidence, box_class_probs, threshold)
36 # same dimension as box_class_scores, and be True for the boxes you want to keep (with probability >= threshold)
37 ## (≈ 1 line)
—> 38 filtering_mask = (box_class_scores >= threshold)
39 # Step 4: Apply the mask to box_class_scores, boxes and box_classes
40 ## (≈ 3 lines)

/opt/conda/lib/python3.7/site-packages/tensorflow/python/ops/gen_math_ops.py in greater_equal(x, y, name)
4053 _result = pywrap_tfe.TFE_Py_FastPathExecute(
4054 _ctx._context_handle, tld.device_name, “GreaterEqual”, name,
→ 4055 tld.op_callbacks, x, y)
4056 return _result
4057 except _core._NotOkStatusException as e:

TypeError: Cannot convert 0.5 to EagerTensor of dtype int64

That means your box_class_scores tensor has the wrong contents down in yolo_filter_boxes. A common mistake is to mix up the definition of box_classes and box_class_scores.

Did you compare your results to what I showed above:

1 Like

Thanks @paulinpaloalto yes your are right i made mistake in box_class_scores i given wrong function, i cleared the test cases , in fact i completed the whole assignment with 100/100.Thanks for the help sir

Hello @paulinpaloalto, I followed some direction from the original post above in order to figure out what might be my own mistake; however, I’m still stuck.

This is what I did: box_classes = tf.math.argmax(box_scores, axis=-1)
box_class_scores = tf.math.reduce_max(box_scores, axis=-1, keepdims=False)
filtering_mask = tf.math.greater_equal(box_class_scores, threshold)

And how I applied the mask to scores: boxes = tf.boolean_mask(scores, filtering_mask)
The rest is pretty similar.
I am getting the following error:


I don’t actually know what I did wrong

Could you please provide an assistance.

Sorry, I figured out what was the problem, I solved it. I didn’t apply the mask well to boxes. I had to wear my glasses.