Error Loading Model

[Robin2091]

I have saved the yolov3 tiny model into HDF5 format. However, when I try to load the model using keras load_model function I get this error.
Traceback (most recent call last):

File "Save_Model.py", line 40, in
new_model = keras.models.load_model('./savedmodel2/YOLO.h5')#tf.saved_model.load_model
File "C:\Users\sudes\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow_core\python\keras\saving\save.py", line 146, in load_model
return hdf5_format.load_model_from_hdf5(filepath, custom_objects, compile)
File "C:\Users\sudes\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow_core\python\keras\saving\hdf5_format.py", line 168, in load_model_from_hdf5
custom_objects=custom_objects)
File "C:\Users\sudes\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow_core\python\keras\saving\model_config.py", line 55, in model_from_config
return deserialize(config, custom_objects=custom_objects)
File "C:\Users\sudes\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow_core\python\keras\layers\serialization.py", line 102, in deserialize
printable_module_name='layer')
File "C:\Users\sudes\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow_core\python\keras\utils\generic_utils.py", line 191, in deserialize_keras_object
list(custom_objects.items())))
File "C:\Users\sudes\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow_core\python\keras\engine\network.py", line 906, in from_config
config, custom_objects)
File "C:\Users\sudes\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow_core\python\keras\engine\network.py", line 1852, in reconstruct_from_config
process_node(layer, node_data)
File "C:\Users\sudes\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow_core\python\keras\engine\network.py", line 1799, in process_node
output_tensors = layer(input_tensors, kwargs)
File "C:\Users\sudes\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow_core\python\keras\engine\base_layer.py", line 842, in call
outputs = call_fn(cast_inputs, *args, kwargs)
File "C:\Users\sudes\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow_core\python\keras\layers\core.py", line 795, in call
return self.function(inputs, arguments)
File "C:/Users/sudes/Documents/yolov3-tf2-master/yolov3-tf2-master/yolov3_tf2/models.py", line 252, in
boxes_0 = Lambda(lambda x: yolo_boxes(x, anchors[masks[0]], classes),
TypeError: list indices must be integers or slices, not list

How can I solve this problem? Is it something with the model itself or is something going wrong when saving in this specific format? I was able to get it to work in the TensorFlow saved model format. But I need it in an hdf5 file.

I also need to know what the shape of the input layer is to the yolov3 tiny model.

Thank you.

[zzh8829]

i did use hdf5 format initially and they don't work for some reason.
tensorflow/tensorflow#27769
keras-team/keras#11847

it look like the old problem was fixed and this one is a new problem with hdf5.
I don't have to much faith in fixing hdf5 because they are not native to tensorflow. Is there a reason you can't use tf format.

For your last question, you can print out model summary using yolo.summary() like convert.py.

[Robin2091]

@zzh8829 Hi, thanks for your help. I was able to find the input dimensions of the model. I am trying to integrate this model into unity and I need either a frozen graph as a .pb file or an h5 file

[zzh8829]

I believe either .tf or .h5 are for model (training) checkpoints. For your purpose SavedModel aka .pb format is more suitable. Please check out tools/export_tfserving.py on how to export model to .pb format

[stanleyyylau]

I believe either .tf or .h5 are for model (training) checkpoints. For your purpose SavedModel aka .pb format is more suitable. Please check out tools/export_tfserving.py on how to export model to .pb format

Hi, I would like to use the model with tfjs, should I convert to .h5 format first than use tfconverter to convert to browser format? which way do you recommend?

[ptamas88]

any progress with this? trying to do the same (load the model in h5 format) to add one more input and output to the model

[edurenye]

I don't get the error when I export the model as .pb, but I get the error when I import the model:

File "/projects/yolov3_tf2/models.py", line 250, in <lambda>
TypeError: list indices must be integers or slices, not list

[edurenye]

We have:

anchors = np.array([(10, 14), (23, 27), (37, 58),
                              (81, 82), (135, 169),  (344, 319)],
                             np.float32) / 416
masks = np.array([[3, 4, 5], [0, 1, 2]])

And we map it like this:

anchors[masks[0]]

If instead I replace anchors[masks[0]] by its values, like:

boxes_0 = Lambda(lambda x: yolo_boxes(x,
                 np.array([(81, 82), (135, 169),  (344, 319)], np.float32) / 416, classes),
                 name='yolo_boxes_0')(output_0)
boxes_1 = Lambda(lambda x: yolo_boxes(x,
                 np.array([(10, 14), (23, 27), (37, 58)], np.float32) / 416, classes),
                 name='yolo_boxes_1')(output_1)

Then it just works but I understand this is not supposed to be like this, works for me as a workaround, but I would like to have a proper solution.

Is somewhere the np.array transformed into a list?

[edurenye]

With that workaround I been able to deploy the model to my Raspberry pi, it loads the model fine, but when I try to to call boxes, scores, classes, nums = self.model(img) it throws the following error:

ValueError: not enough values to unpack (expected 4, got 2)

Looks like are 2 tensos:
First: <tf.Tensor: shape=(1, 13, 13, 3, 8), dtype=float32, numpy=array(....), dtype=float32)>
Second: <tf.Tensor: shape=(1, 26, 26, 3, 8), dtype=float32, numpy=array(....), dtype=float32)>

The 3 in there is probably because I'm using 3 custom classes.

Any idea why this happens?