Output channels in UNet colab

Course Q&A TensorFlow: Advanced Techniques Specialization Advanced Computer Vision with TensorFlow
1

Hi, I have a question,
why did we not convert the UNet output (segments) to one hot encoding,
we configure the output to have 3 channels but the training set has 1 channel and it seems like the fit function does not have a problem with it. Does model.fit convert to one hot enconding automatically?

2

The train_dataset does not have 1 channel. It is a tensor of 4 dimensions i.e. 3 channels and batch size.

3

I only see 1 channel:

train = dataset[‘train’].map(load_image_train, num_parallel_calls=tf.data.experimental.AUTOTUNE)

test = dataset[‘test’].map(load_image_test)

print(train)

for a, b in train:

print(b.shape)

break

<ParallelMapDataset shapes: ((128, 128, 3), (128, 128, 1)), types: (tf.float32, tf.uint8)>
(128, 128, 1)

4

Yes you are right input is (None, 64,84, 1) my mistake, and output is (None, 64, 84, 11) . From what I see the entire model converts the input to this output. The model.fit will fit as per the model. This is the week 3 assignment.

1 Like
5

Thanks for confirming :slight_smile:

1 Like