InvalidArgumentError: required broadcastable shapes [Op:Mul]

General Discussion
, ,
#32

Thanks for the quick reply. I’m looking forward to the fix.

#34

find the object_detector_spec.py in anaconda3/envs/tf2.5/lib/python3.6/site-packages/tensorflow_examples/lite/model_maker/core/task/model_spec, then change nms_boxes, nms_classes, nms_scores, _ = lite_runner.run(images) ----> nms_scores, nms_boxes, nms_count, nms_classes = lite_runner.run(images), should address the error in tf2.6-gpu

#36

Is there any update about the issue, please? Maybe can you let us know when it going to be fixing exactly…

#37

Sorry that it takes longer time than we expected. Will notify you here once it’s released.

1 Like
#39

Any new update on this issue?

1 Like
#41

We have released a new version of model maker.

1 Like
#43

Thanks a lot for the hard work. I’ll try it out as soon as possible.

#44

My quick test with the salad maker colab demo failed at " Step 3. Train the TensorFlow model with the training data."

I also tried with modelmaker nightly.

UnknownError: 2 root error(s) found.
(0) Unknown: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above.
[[{{node keras_layer/StatefulPartitionedCall/StatefulPartitionedCall/StatefulPartitionedCall/efficientnet-lite2/StatefulPartitionedCall/stem/conv2d/Conv2D}}]]
(1) Unknown: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above.
[[{{node keras_layer/StatefulPartitionedCall/StatefulPartitionedCall/StatefulPartitionedCall/efficientnet-lite2/StatefulPartitionedCall/stem/conv2d/Conv2D}}]]
[[Func/batch_cls_loss/write_summary/summary_cond/then/_4448/input/_8959/_96]]
0 successful operations.
0 derived errors ignored. [Op:__inference_train_function_126022]

#45

fixed this by switching to high memory instance.

#46

Well it ran almost all the way through the colab example. But in the end when it tries to test the model it’s failing.

#@title Run object detection and show the detection results

gives me:

0.3
<class ‘float’>

IndexError Traceback (most recent call last)
in ()
20 TEMP_FILE,
21 interpreter,
—> 22 threshold=DETECTION_THRESHOLD
23 )
24

1 frames
in detect_objects(interpreter, image, threshold)
61 results = []
62 for i in range(count):
—> 63 if scores[i] >= threshold:
64 result = {
65 ‘bounding_box’: boxes[i],

IndexError: too many indices for array: array is 0-dimensional, but 1 were indexed

1 Like
#48

This error occurs because the output format has changed. Replacing

boxes = get_output_tensor(interpreter, 0)
classes = get_output_tensor(interpreter, 1)
scores = get_output_tensor(interpreter, 2)
count = int(get_output_tensor(interpreter, 3))

with:

scores = get_output_tensor(interpreter, 0)
boxes = get_output_tensor(interpreter, 1)
count = int(get_output_tensor(interpreter, 2))
classes = get_output_tensor(interpreter, 3)

I’ll create a PR to fix this in the Train_a_salad_detector_with_TFLite_Model_Maker.ipynb notebook.

Kind regards,
Gilbert Tanner

#49

Thank you, I’m trying to get this to work now.

#51

I updated the Train_a_salad_detector_with_TFLite_Model_Maker.ipynb notebook in this PR. It seems to work with the Android example as well, even though that needs to be further tested.

#52

Oh that’s great, that was my next move. I was able to get it to work on colab. Oddly it failed most of the time when using a GPU instance. But it ran okay with a TPU instance.

#53

Been following this thread and it finally worked for me. Here is what I did …
In the first cell I did NOT use tensorflow 2.5
!pip install -q tflite-model-maker

!pip install -q pycocotools
!pip install -q tflite-support

AND in the second to last cell there was a missing import for image processing
after tflite metadata support I added the image support

from tflite_support import metadata
from PIL import Image

Thats it and it was pretty fast in training and evaluation

#54

I tried to use a copy of the tflite model from the updated notebook in Android Object detection app. … and it did not work. Do you know how to change the order of the outputs in the model… in android object detection app???

#56

Has anyone got the salad colab tflite file to work with android object detection???

#58

This worked for me, I was so confused for so long about what I was doing wrong from the sample code!!