Mask_rcnn: RuntimeError

[Telang8100]

File "C:\Users\SIDDHESHWARAnaconda3\lib\site-packages\keras\engine\network.py", line 317, in __setattr__
'It looks like you are subclassing Model and you '

RuntimeError: It looks like you are subclassing Model and you forgot to call super(YourClass, self).__init__(). Always start with this line.

[houzhizun]

same problem!!!!!
when I want to train by mult-GPU,
RuntimeError: It looks like you are subclassing Model and you forgot to call super(YourClass, self).__init__(). Always start with this line.

[sumeet70]

Same issue here as well.
When GPU_COUNT > 1

[decewei]

pip3 install keras==2.1.3. This solves the issue for training, you can train on multi GPU. It won't work for inference, because keras don't have savings in 2.1.3 build.

[decewei]

Solved by changing code in mrcnn/parallel_model.py as the following:

class ParallelModel(KM.Model):
    def __init__(self, keras_model, gpu_count):
        """Class constructor.
        keras_model: The Keras model to parallelize
        gpu_count: Number of GPUs. Must be > 1
        """
        super(ParallelModel, self).__init__()
        self.inner_model = keras_model
        self.gpu_count = gpu_count
        merged_outputs = self.make_parallel()
        super(ParallelModel, self).__init__(inputs=self.inner_model.inputs,
                                            outputs=merged_outputs)

[shen338]

Solved by changing code in mrcnn/parallel_model.py as the following:

class ParallelModel(KM.Model):
    def __init__(self, keras_model, gpu_count):
        """Class constructor.
        keras_model: The Keras model to parallelize
        gpu_count: Number of GPUs. Must be > 1
        """
        super(ParallelModel, self).__init__()
        self.inner_model = keras_model
        self.gpu_count = gpu_count
        merged_outputs = self.make_parallel()
        super(ParallelModel, self).__init__(inputs=self.inner_model.inputs,
                                            outputs=merged_outputs)

When change code in your way, It still has error, asking for two arguments: inputs and outputs

[shen338]

Solved by changing code in mrcnn/parallel_model.py as the following:

class ParallelModel(KM.Model):
    def __init__(self, keras_model, gpu_count):
        """Class constructor.
        keras_model: The Keras model to parallelize
        gpu_count: Number of GPUs. Must be > 1
        """
        super(ParallelModel, self).__init__()
        self.inner_model = keras_model
        self.gpu_count = gpu_count
        merged_outputs = self.make_parallel()
        super(ParallelModel, self).__init__(inputs=self.inner_model.inputs,
                                            outputs=merged_outputs)

When change code in your way, It still has error, asking for two arguments: inputs and outputs

Sorry, It works. My bad. Thank you!

[zhouhao-learning]

@shen338
When I think you changed the above, I got an error, suggesting that I need two parameters, one input, one output, how do you solve this error?

[VinayChaudhari1996]

I got the result of training only does a single change :
Step 1 :
pip3 install keras==2.1.3.

Step 2:
Don't change any line of code and run the file again!

---

I got good accuracy 👍

[hachreak]

Solved by changing code in mrcnn/parallel_model.py as the following:

class ParallelModel(KM.Model):
    def __init__(self, keras_model, gpu_count):
        """Class constructor.
        keras_model: The Keras model to parallelize
        gpu_count: Number of GPUs. Must be > 1
        """
        super(ParallelModel, self).__init__()
        self.inner_model = keras_model
        self.gpu_count = gpu_count
        merged_outputs = self.make_parallel()
        super(ParallelModel, self).__init__(inputs=self.inner_model.inputs,
                                            outputs=merged_outputs)

The trick works also for me... :smile:
Keras v2.2.4

[vishwakarmarhl]

This solution works with "Keras==2.2.2"

[SevenFoldDeep]

@shen338 You say you ran into the error of requiring input and output, how did you fix this???

[AbhinayReddyYarva]

Solved by changing code in mrcnn/parallel_model.py as the following:

class ParallelModel(KM.Model):
    def __init__(self, keras_model, gpu_count):
        """Class constructor.
        keras_model: The Keras model to parallelize
        gpu_count: Number of GPUs. Must be > 1
        """
        super(ParallelModel, self).__init__()
        self.inner_model = keras_model
        self.gpu_count = gpu_count
        merged_outputs = self.make_parallel()
        super(ParallelModel, self).__init__(inputs=self.inner_model.inputs,
                                            outputs=merged_outputs)

Thanks, above change accepts multiple GPUs count but getting below error while training on multiple GPU's

Starting at epoch 0. LR=0.001

Checkpoint Path: ../matterport_mask_rcnn/logs/cocosynth_dataset20191125T0544/mask_rcnn_cocosynth_dataset_{epoch:04d}.h5
Selecting layers to train
fpn_c5p5 (Conv2D)
fpn_c4p4 (Conv2D)
fpn_c3p3 (Conv2D)
fpn_c2p2 (Conv2D)
fpn_p5 (Conv2D)
fpn_p2 (Conv2D)
fpn_p3 (Conv2D)
fpn_p4 (Conv2D)
In model: rpn_model
rpn_conv_shared (Conv2D)
rpn_class_raw (Conv2D)
rpn_bbox_pred (Conv2D)
mrcnn_mask_conv1 (TimeDistributed)
mrcnn_mask_bn1 (TimeDistributed)
mrcnn_mask_conv2 (TimeDistributed)
mrcnn_mask_bn2 (TimeDistributed)
mrcnn_class_conv1 (TimeDistributed)
mrcnn_class_bn1 (TimeDistributed)
mrcnn_mask_conv3 (TimeDistributed)
mrcnn_mask_bn3 (TimeDistributed)
mrcnn_class_conv2 (TimeDistributed)
mrcnn_class_bn2 (TimeDistributed)
mrcnn_mask_conv4 (TimeDistributed)
mrcnn_mask_bn4 (TimeDistributed)
mrcnn_bbox_fc (TimeDistributed)
mrcnn_mask_deconv (TimeDistributed)
mrcnn_class_logits (TimeDistributed)

mrcnn_mask (TimeDistributed)

AttributeError Traceback (most recent call last)
in
7 learning_rate=config.LEARNING_RATE,
8 epochs=4,
----> 9 layers='heads')
10 end_train = time.time()
11 minutes = round((end_train - start_train) / 60, 2)

~/objectDetection/matterport_mask_rcnn/mrcnn/model.py in train(self, train_dataset, val_dataset, learning_rate, epochs, layers, augmentation, custom_callbacks, no_augmentation_sources)
2352 log("Checkpoint Path: {}".format(self.checkpoint_path))
2353 self.set_trainable(layers)
-> 2354 self.compile(learning_rate, self.config.LEARNING_MOMENTUM)
2355
2356 # Work-around for Windows: Keras fails on Windows when using

~/objectDetection/matterport_mask_rcnn/mrcnn/model.py in compile(self, learning_rate, momentum)
2197 tf.reduce_mean(layer.output, keepdims=True)
2198 * self.config.LOSS_WEIGHTS.get(name, 1.))
-> 2199 self.keras_model.metrics_tensors.append(loss)
2200
2201 def set_trainable(self, layer_regex, keras_model=None, indent=0, verbose=1):

~/objectDetection/matterport_mask_rcnn/mrcnn/parallel_model.py in __getattribute__(self, attrname)
45 if 'load' in attrname or 'save' in attrname:
46 return getattr(self.inner_model, attrname)
---> 47 return super(ParallelModel, self).__getattribute__(attrname)
48
49 def summary(self, args, *kwargs):

AttributeError: 'ParallelModel' object has no attribute 'metrics_tensors'