ValueError: Tried to convert 'shape' to a tensor and failed. After TensorFlow updating.

[kiflowb777]

After update TensorFlow 1.14 to 2.0 and use tf.keras instead of keras, when using fpn_classifier_graph I get:

ValueError: Tried to convert 'shape' to a tensor and failed. Error: None values not supported.

  File "D:\project\project.py", line 55, in fpn_classifier_graph
    mrcnn_bbox = KL.Reshape((s[1], num_classes, 4), name="mrcnn_bbox")(x)
  File "C:\python36\lib\site-packages\tensorflow\python\keras\engine\base_layer.py", line 634, in __call__
    outputs = call_fn(inputs, *args, **kwargs)
  File "C:\python36\lib\site-packages\tensorflow\python\keras\layers\core.py", line 467, in call
    (array_ops.shape(inputs)[0],) + self.target_shape)
  File "C:\python36\lib\site-packages\tensorflow\python\ops\gen_array_ops.py", line 9094, in reshape
    "Reshape", tensor=tensor, shape=shape, name=name)
  File "C:\python36\lib\site-packages\tensorflow\python\framework\op_def_library.py", line 545, in _apply_op_helper
    (input_name, err))
ValueError: Tried to convert 'shape' to a tensor and failed. Error: None values not supported.

[sainatarajan]

The value is "None". Make sure that in this line: mrcnn_bbox = KL.Reshape((s[1], num_classes, 4), name="mrcnn_bbox")(x) s[1] is not None or maybe x.

[kiflowb777]

Thanks.

With Keras 2.2.4 and TensorFlow 1.14 s is tuple where s[1] = 1000 (int) and x = Tensor("mrcnn_bbox_fc/Reshape_1:0", shape=(?, 1000, 20), dtype=float32)

With tf.keras (2.2.4-tf) and TensorFlow (both 1.14 and 2.0) s[1] = None and x = Tensor("mrcnn_bbox_fc/Reshape_1:0", shape=(?, ?, 20), dtype=float32)

[sainatarajan]

@kiflowb777 Great find! The problem is with the version issues then.

[kiflowb777]

All inputs in both Keras and tf.keras are the same in the fpn_classifier_graph() function. But I noticed that calling the internal function: x = KL.TimeDistributed(KL.Conv2D(fc_layers_size, (pool_size, pool_size), padding="valid"),name="mrcnn_class_conv1")(x) gives a different result: (1.) x = Tensor("mrcnn_class_conv1/Reshape_1:0", shape=(?, 1000, 1, 1, 256), dtype=float32) (2.) x = Tensor("mrcnn_class_conv1/Reshape_1:0", shape=(?, ?, 1, 1, 256), dtype=float32)

(1.) is Keras, (2.) is tf.keras (both tf_version 1.14).

[kiflowb777]

I noticed that in tf.keras in function fpn_classifier_graph the input rois does not have the parameter _keras_shape that contains a num_rois (which in my case has a value of 1000). In my project the rois are returned from ProposalLayer.

I found a similar problem: https://stackoverflow.com/questions/54613474/tensorflow-keras-input-layer-does-not-add-keras-shape

[sainatarajan]

@kiflowb777 I think you should post this on the Tensorflow issues page. This seems like some bug on their part and because they would not have adopted this recent change in their tf.keras package. So, this is working for you when you use keras and not tf.keras right?

[kiflowb777]

I posted this on TensorFlow page: https://github.com/tensorflow/tensorflow/issues/33785

I also noticed that in Keras the overwritten function compute_output_shape is called, but not in the tf.keras.

For this reason, the shape is miscalculated by the ProposalLayer.

[kiflowb777]

I temporarily fixed it by adding a line: rois.set_shape((None, num_rois, 4)) at the beginning of the function: fpn_classifier_graph but during training:

   File "C:\Project\mrcnn\model.py", line 2199, in compile
    self.keras_model.metrics_tensors.append(loss)
AttributeError: 'Model' object has no attribute 'metrics_tensors'

[sainatarajan]

This is definitely a Keras/TensorFlow version issue. self.keras_model is a Keras model object and maybe in the old/new version of Keras, they might have removed it I guess. I am not sure.

[kiflowb777]

The AttributeError: 'Model' object has no attribute 'metrics_tensors' can by fixed by changing line model.metrics_tensors.append(loss) to model.metrics.append(loss)

but then I get:

in get_layer_elements_count
    input_shapes = layer.input_shape if type(layer.input_shape) is list else [layer.input_shape]
  File "C:\python\lib\site-packages\tensorflow\python\keras\engine\base_layer.py", line 1391, in input_shape
    ' has multiple inbound nodes, '
AttributeError: The layer "rpn_model has multiple inbound nodes, with different input shapes. Hence the notion of "input shape" is ill-defined for the layer. Use `get_input_shape_at(node_index)` instead.

[paka0401]

I had the same error AttributeError: 'Model' object has no attribute 'metrics_tensors. I fixed it by downgrading keras to 2.1.0.

By the way I also had to downgrade numpy to 1.16.0 and tensorflow to 1.15.0 to be able to use the Mask_RCNN library. The code I am now able to run can be found here: https://machinelearningmastery.com/how-to-train-an-object-detection-model-with-keras/

[K-M-Ibrahim-Khalilullah]

@kiflowb777 Have yo solved the problem. I have the same problem on this line: mrcnn_bbox = KL.Reshape((s[1], num_classes, 4), name="mrcnn_bbox")(x) My tensorflow version: 1.13.1

Is there any suggestion please? thanks

[kiflowb777]

@ibrahimLearning
Something they started fixing in: https://github.com/tensorflow/tensorflow/pull/34248 You can try run your code with tf-nightly: https://pypi.org/project/tf-nightly-gpu/ Maybe in TF2.1 the shape issue will be fixed.

[kiflowb777]

You can also try call tf.reshape(out_tensor, self.compute_output_shape(input_shape)) after each usage of Custom Layer. More info: https://stackoverflow.com/questions/51028861/tensorflow-compute-output-shape-not-working-for-custom-layer https://github.com/tensorflow/tensorflow/issues/33785

But in spite of this, the model summary differ in keras and tf.keras.

[lanternx]

@ibrahimLearning Something they started fixing in: tensorflow/tensorflow#34248 You can try run your code with tf-nightly: https://pypi.org/project/tf-nightly-gpu/ Maybe in TF2.1 the shape issue will be fixed.

I wonder how to solve this problem? My tensorflow version is 2.2.

[nguyendinh1987]

To solve the problem, you can change this line: mrcnn_bbox = KL.Reshape((s[1], num_classes, 4), name="mrcnn_bbox")(x) by : mrcnn_bbox = KL.Reshape((-1, num_classes, 4), name="mrcnn_bbox")(x)

It works for me and the output is fine.

[dji-mmmjukimo]

To solve the problem, you can change this line: mrcnn_bbox = KL.Reshape((s[1], num_classes, 4), name="mrcnn_bbox")(x) by : mrcnn_bbox = KL.Reshape((-1, num_classes, 4), name="mrcnn_bbox")(x)

It works for me and the output is fine.

@nguyendinh1987 I tried your way, but the error still exist: ValueError: Tried to convert 'shape' to a tensor and failed. Error: None values not supported. My tf vision is 2.3.0 and keras vision is 2.4.3 Do you know any model.py file can solve the all different vision problem?

[nguyendinh1987]

@dji-mmmjukimo : My tf and Keras version are 1.13.0-rc and 2.2.4-tf respectively. According to tf.keras.Reshape API for your tf version (https://www.tensorflow.org/api_docs/python/tf/keras/layers/Reshape) input -1 is accepted. I think the problem is not because of -1 or s[1].

---

ValueError: Tried to convert 'shape' to a tensor and failed. Error: None values not supported.

In the error message, "None values not supported" might stand for one of your input values fed to Reshape function is None. Can you try to check value of variables "num_classes" and "x" when you run your program.

---

Do you know any model.py file can solve the all different vision problem?

I do not have any idea :). Each version has its own specification and we need to modify our code to fit with it.

[lanternx]

@dji-mmmjukimo : My tf and Keras version are 1.13.0-rc and 2.2.4-tf respectively. According to tf.keras.Reshape API for your tf version (https://www.tensorflow.org/api_docs/python/tf/keras/layers/Reshape) input -1 is accepted. I think the problem is not because of -1 or s[1].

ValueError: Tried to convert 'shape' to a tensor and failed. Error: None values not supported.

In the error message, "None values not supported" might stand for one of your input values fed to Reshape function is None. Can you try to check value of variables "num_classes" and "x" when you run your program.

Do you know any model.py file can solve the all different vision problem?

I do not have any idea :). Each version has its own specification and we need to modify our code to fit with it.

Thank you. But seems hard for an unskilled coder like me. :-|

[Al-Badri179]

KL.Reshape

The value in not none and the same problem is exist

[kiflowb777]

@ML-BOSS Please try this:

if s[1] is None:
    mrcnn_bbox = KL.Reshape((-1, num_classes, 4), name="mrcnn_bbox")(x)
else:
    mrcnn_bbox = KL.Reshape((s[1], num_classes, 4), name="mrcnn_bbox")(x)

[Al-Badri179]

if s[1] is None: mrcnn_bbox = KL.Reshape((-1, num_classes, 4), name="mrcnn_bbox")(x) else: mrcnn_bbox = KL.Reshape((s[1], num_classes, 4), name="mrcnn_bbox")(x)

I tried it and the problem didn't solve. I have this error:

File "c:\users\msi-pc\appdata\local\programs\python\python37\lib\site-packages\mask_rcnn-2.1-py3.7.egg\mrcnn\model.py", line 954 mrcnn_bbox = layers.Reshape((-1, num_classes, 4), name="mrcnn_bbox")(x) ^ IndentationError: expected an indented block

[kiflowb777]

@Al-Badri179 which version of the TensorFlow are you using?

[rampam]

if s[1] is None: mrcnn_bbox = KL.Reshape((-1, num_classes, 4), name="mrcnn_bbox")(x) else: mrcnn_bbox = KL.Reshape((s[1], num_classes, 4), name="mrcnn_bbox")(x)

I tried it and the problem didn't solve. I have this error:

File "c:\users\msi-pc\appdata\local\programs\python\python37\lib\site-packages\mask_rcnn-2.1-py3.7.egg\mrcnn\model.py", line 954 mrcnn_bbox = layers.Reshape((-1, num_classes, 4), name="mrcnn_bbox")(x) ^ IndentationError: expected an indented block

You forgot the indentation after the if and else statements.

[Liuxuweilxw]

Try to uninstall keras and tensorflow ,then install tensorflow==1.3.0 and keras==2.0.8. If you don't want to uninstall tensorflow2.x which has been installed on your system，try to create a new environment.

[madhumnk]

---

ValueError Traceback (most recent call last) ~\AppData\Local\Temp\ipykernel_9028\2113197888.py in 1 print("loading weights for Mask R-CNN model…") ----> 2 model = modellib.MaskRCNN(mode="inference", config=config, model_dir='./')

~\Mask_RCNN\mrcnn\model.py in init(self, mode, config, model_dir) 1842 self.model_dir = model_dir 1843 self.set_log_dir() -> 1844 self.keras_model = self.build(mode=mode, config=config) 1845 1846 def build(self, mode, config):

~\Mask_RCNN\mrcnn\model.py in build(self, mode, config) 2036 # Network Heads 2037 # Proposal classifier and BBox regressor heads -> 2038 mrcnn_class_logits, mrcnn_class, mrcnn_bbox = fpn_classifier_graph(rpn_rois, mrcnn_feature_maps, input_image_meta, 2039 config.POOL_SIZE, config.NUM_CLASSES, 2040 train_bn=config.TRAIN_BN,

~\Mask_RCNN\mrcnn\model.py in fpn_classifier_graph(rois, feature_maps, image_meta, pool_size, num_classes, train_bn, fc_layers_size) 964 # Reshape to [batch, num_rois, NUM_CLASSES, (dy, dx, log(dh), log(dw))] 965 s = K.int_shape(x) --> 966 mrcnn_bbox = KL.Reshape((s[1], num_classes, 4), name="mrcnn_bbox")(x) 967 968 return mrcnn_class_logits, mrcnn_probs, mrcnn_bbox

~\anaconda3\lib\site-packages\keras\engine\base_layer_v1.py in call(self, *args, *kwargs) 836 self._compute_dtype_object 837 ): --> 838 outputs = call_fn(cast_inputs, args, **kwargs) 839 840 except tf.errors.OperatorNotAllowedInGraphError as e:

~\anaconda3\lib\site-packages\keras\layers\reshaping\reshape.py in call(self, inputs) 135 136 def call(self, inputs): --> 137 result = tf.reshape(inputs, (tf.shape(inputs)[0],) + self.target_shape) 138 if not tf.executing_eagerly(): 139 # Set the static shape for the result since it might lost during

~\AppData\Roaming\Python\Python39\site-packages\tensorflow\python\util\traceback_utils.py in error_handler(*args, **kwargs) 151 except Exception as e: 152 filtered_tb = _process_traceback_frames(e.traceback) --> 153 raise e.with_traceback(filtered_tb) from None 154 finally: 155 del filtered_tb

~\AppData\Roaming\Python\Python39\site-packages\tensorflow\python\framework\op_def_library.py in _ExtractInputsAndAttrs(op_type_name, op_def, allowed_list_attr_map, keywords, default_type_attr_map, attrs, inputs, input_types) 569 values, as_ref=input_arg.is_ref).dtype.name 570 except ValueError as err: --> 571 raise ValueError( 572 f"Tried to convert '{input_name}' to a tensor and failed. " 573 f"Error: {err}")

ValueError: Tried to convert 'shape' to a tensor and failed. Error: None values not supported.

[madhumnk]

ValueError Traceback (most recent call last) ~\AppData\Local\Temp\ipykernel_9028\2113197888.py in 1 print("loading weights for Mask R-CNN model…") ----> 2 model = modellib.MaskRCNN(mode="inference", config=config, model_dir='./')

~\Mask_RCNN\mrcnn\model.py in init(self, mode, config, model_dir) 1842 self.model_dir = model_dir 1843 self.set_log_dir() -> 1844 self.keras_model = self.build(mode=mode, config=config) 1845 1846 def build(self, mode, config):

~\Mask_RCNN\mrcnn\model.py in build(self, mode, config) 2036 # Network Heads 2037 # Proposal classifier and BBox regressor heads -> 2038 mrcnn_class_logits, mrcnn_class, mrcnn_bbox = fpn_classifier_graph(rpn_rois, mrcnn_feature_maps, input_image_meta, 2039 config.POOL_SIZE, config.NUM_CLASSES, 2040 train_bn=config.TRAIN_BN,

~\Mask_RCNN\mrcnn\model.py in fpn_classifier_graph(rois, feature_maps, image_meta, pool_size, num_classes, train_bn, fc_layers_size) 964 # Reshape to [batch, num_rois, NUM_CLASSES, (dy, dx, log(dh), log(dw))] 965 s = K.int_shape(x) --> 966 mrcnn_bbox = KL.Reshape((s[1], num_classes, 4), name="mrcnn_bbox")(x) 967 968 return mrcnn_class_logits, mrcnn_probs, mrcnn_bbox

~\anaconda3\lib\site-packages\keras\engine\base_layer_v1.py in call(self, *args, *kwargs) 836 self._compute_dtype_object 837 ): --> 838 outputs = call_fn(cast_inputs, args, **kwargs) 839 840 except tf.errors.OperatorNotAllowedInGraphError as e:

~\anaconda3\lib\site-packages\keras\layers\reshaping\reshape.py in call(self, inputs) 135 136 def call(self, inputs): --> 137 result = tf.reshape(inputs, (tf.shape(inputs)[0],) + self.target_shape) 138 if not tf.executing_eagerly(): 139 # Set the static shape for the result since it might lost during

~\AppData\Roaming\Python\Python39\site-packages\tensorflow\python\util\traceback_utils.py in error_handler(*args, kwargs) 151 except Exception as e: 152 filtered_tb = _process_traceback_frames(e.traceback**) --> 153 raise e.with_traceback(filtered_tb) from None 154 finally: 155 del filtered_tb

~\AppData\Roaming\Python\Python39\site-packages\tensorflow\python\framework\op_def_library.py in _ExtractInputsAndAttrs(op_type_name, op_def, allowed_list_attr_map, keywords, default_type_attr_map, attrs, inputs, input_types) 569 values, as_ref=input_arg.is_ref).dtype.name 570 except ValueError as err: --> 571 raise ValueError( 572 f"Tried to convert '{input_name}' to a tensor and failed. " 573 f"Error: {err}")

ValueError: Tried to convert 'shape' to a tensor and failed. Error: None values not supported.

Getting error after trying all sort of solutions provided here, Please help me with it