Add to the Loaders get_batch function the possibility of loading it in another process.

If its the first time to call it, load a batch as normal and creates anew process to load next batch

it is not the first time to load it, creates a new process to load next batch and waits for the previous loaded batch in the queue.

https://stackoverflow.com/questions/2046603/is-it-possible-to-run-function-in-a-subprocess-without-threading-or-writing-a-se

compare both performances

Hi,

I change the example to use my dataset. I have just one class (the mask has two colors, black and with). The image has 140x140 pixels.

n_classes = 1 dataset_path = 'seg' loader = Loader.Loader(dataFolderPath=dataset_path, n_classes=n_classes, problemType='segmentation', width=140, height=140, ignore_label=n_classes) model = MnasnetEager.MnasnetFC(num_classes=n_classes) optimizer = tf.train.AdamOptimizer(0.001) train(loader=loader, model=model, epochs=20, batch_size=8) get_params(model)

The result generates an error: Using TensorFlow backend. Reading files... Structuring test and train files... Loaded 92 training samples Loaded 10 testing samples Dataset contains 1 classes epoch: 0 Traceback (most recent call last): File "train_eager_seg.py", line 109, in train(loader=loader, model=model, epochs=20, batch_size=1) File "train_eager_seg.py", line 44, in train x, y, mask = loader.get_batch(size=batch_size, train=True, augmenter=augmenter) File "/pylon5/ac3uump/rafaelmr/Semantic-Segmentation-Tensorflow-Eager/Loader.py", line 377, in get_batch return self._get_batch_segmentation(size=size, train=train, augmenter=augmenter) File "/pylon5/ac3uump/rafaelmr/Semantic-Segmentation-Tensorflow-Eager/Loader.py", line 296, in _get_batch_segmentation y = to_categorical(y, num_classes=self.n_classes+1) File "/home/rafaelmr/.conda/envs/seg1/lib/python2.7/site-packages/keras/utils/np_utils.py", line 32, in to_categorical categorical[np.arange(n), y] = 1 IndexError: index 255 is out of bounds for axis 1 with size 2

Probably the error is associated with the number of the classes. I put the correct number of the classes. I try others numbers, but the problem still the same. Do I need to code another place?

Are the deconv (transpose conv) going to create artifacts in the decoder?

https://distill.pub/2016/deconv-checkerboard/ (there are many mentions to this 2016 paper)

First of all, thank you for sharing. I use the original dataset to get the correct results, but the final predicted result image of the cityscapes dataset using RGB annotations is all black, and the output is black when I test with my own dataset. Do you know why? thank you very much!

Hi,

I executed your code, but the results are strange: epoch: 0 Train accuracy: None Test accuracy: None epoch: 1 Train accuracy: None Test accuracy: None ...... epoch: 18 Train accuracy: None Test accuracy: None epoch: 19 Train accuracy: None Test accuracy: None Total parameters of the net: 6421507 == 6.421507M

Why are all accuracies with none result?

Tks

Hello, thanks for your work with this code. I found a lot of class of network in the Network.py. Can you tell me those networks built depending on the existing the network structure or designed by yourself?

I have never used tensorflow eager, and I want to run the saved model on the mobile phone. May I ask you if the model saved by tensorflow eager can be saved as a file in tflite format and run on the mobile phone? Save the model without. What about the meta file? Thank you very much for your help!

Change the custom loader your are using for the Dataset API. Keep the same functionalities like data augmentation. Change the rest of the code to integrate it. https://www.tensorflow.org/tutorials/eager/eager_basics#datasets https://www.tensorflow.org/performance/datasets_performance https://colab.research.google.com/github/tensorflow/tensorflow/blob/master/tensorflow/contrib/eager/python/examples/generative_examples/image_captioning_with_attention.ipynb

Once the Dataset API is integrated, move into the TFrecords data format http://warmspringwinds.github.io/tensorflow/tf-slim/2016/12/21/tfrecords-guide/