Is there a tutorial or documentation for running custom tensorflow network in movidius ?

[idata]

Hello,

I am trying to run facenet (Inception Resnet V1 on movidius). Please point me towards any tutorial as to how to start compiling for tensorflow. I tried mvCCompile on mnist softmax.py example from the tensorflow website and got the following error. Please help. I am also attaching the softmax.py which I used to create a frozen graph.mvNCCompile mnist_output/mnist_default.meta -in=input -s12 -o=mnist_graph -on=output mvNCCompile v02.00, Copyright @ Movidius Ltd 2016 /usr/local/lib/python3.5/dist-packages/tensorflow/python/util/tf_inspect.py:45: DeprecationWarning: inspect.getargspec() is deprecated, use inspect.signature() instead if d.decorator_argspec is not None), _inspect.getargspec(target)) [Error 34] Setup Error: Values for input contain placeholder. Pass an absolute value.

softmax.py code below:

from __future__ import absolute_import from __future__ import division from __future__ import print_function import argparse import sys import tempfile from tensorflow.examples.tutorials.mnist import input_data import tensorflow as tf FLAGS = None def deepnn(x): """deepnn builds the graph for a deep net for classifying digits. Args: x: an input tensor with the dimensions (N_examples, 784), where 784 is the number of pixels in a standard MNIST image. Returns: A tuple (y, keep_prob). y is a tensor of shape (N_examples, 10), with values equal to the logits of classifying the digit into one of 10 classes (the digits 0-9). keep_prob is a scalar placeholder for the probability of dropout. """ # Reshape to use within a convolutional neural net. # Last dimension is for "features" - there is only one here, since images are # grayscale -- it would be 3 for an RGB image, 4 for RGBA, etc. with tf.name_scope('reshape'): x_image = tf.reshape(x, [-1, 28, 28, 1]) # First convolutional layer - maps one grayscale image to 32 feature maps. with tf.name_scope('conv1'): W_conv1 = weight_variable([5, 5, 1, 32]) b_conv1 = bias_variable([32]) h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1) # Pooling layer - downsamples by 2X. with tf.name_scope('pool1'): h_pool1 = max_pool_2x2(h_conv1) # Second convolutional layer -- maps 32 feature maps to 64. with tf.name_scope('conv2'): W_conv2 = weight_variable([5, 5, 32, 64]) b_conv2 = bias_variable([64]) h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2) # Second pooling layer. with tf.name_scope('pool2'): h_pool2 = max_pool_2x2(h_conv2) # Fully connected layer 1 -- after 2 round of downsampling, our 28x28 image # is down to 7x7x64 feature maps -- maps this to 1024 features. with tf.name_scope('fc1'): W_fc1 = weight_variable([7 * 7 * 64, 1024]) b_fc1 = bias_variable([1024]) h_pool2_flat = tf.reshape(h_pool2, [-1, 7*7*64]) h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1) # Dropout - controls the complexity of the model, prevents co-adaptation of # features. with tf.name_scope('dropout'): keep_prob = tf.placeholder(tf.float32) h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob) # Map the 1024 features to 10 classes, one for each digit with tf.name_scope('fc2'): W_fc2 = weight_variable([1024, 10]) b_fc2 = bias_variable([10]) y_conv = tf.matmul(h_fc1_drop, W_fc2) + b_fc2 return y_conv, keep_prob def conv2d(x, W): """conv2d returns a 2d convolution layer with full stride.""" return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME') def max_pool_2x2(x): """max_pool_2x2 downsamples a feature map by 2X.""" return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME') def weight_variable(shape): """weight_variable generates a weight variable of a given shape.""" initial = tf.truncated_normal(shape, stddev=0.1) return tf.Variable(initial) def bias_variable(shape): """bias_variable generates a bias variable of a given shape.""" initial = tf.constant(0.1, shape=shape) return tf.Variable(initial) def main(_): # Import data mnist = input_data.read_data_sets(FLAGS.data_dir, one_hot=True) # Create the model x = tf.placeholder(tf.float32, [None, 784], name ="input") # Define loss and optimizer y_ = tf.placeholder(tf.float32, [None, 10], name = "output") # Build the graph for the deep net y_conv, keep_prob = deepnn(x) with tf.name_scope('loss'): cross_entropy = tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y_conv) cross_entropy = tf.reduce_mean(cross_entropy) with tf.name_scope('adam_optimizer'): train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy) with tf.name_scope('accuracy'): correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y_, 1)) correct_prediction = tf.cast(correct_prediction, tf.float32) accuracy = tf.reduce_mean(correct_prediction) graph_location = tempfile.mkdtemp() print('Saving graph to: %s' % graph_location) train_writer = tf.summary.FileWriter(graph_location) train_writer.add_graph(tf.get_default_graph()) with tf.Session() as sess: sess.run(tf.global_variables_initializer()) for i in range(20000): batch = mnist.train.next_batch(50) if i % 100 == 0: train_accuracy = accuracy.eval(feed_dict={ x: batch[0], y_: batch[1], keep_prob: 1.0}) print('step %d, training accuracy %g' % (i, train_accuracy)) train_step.run(feed_dict={x: batch[0], y_: batch[1], keep_prob: 0.5}) saver = tf.train.Saver(tf.global_variables()) saver.save(sess, "mnist_output/"+'mnist_default') print('test accuracy %g' % accuracy.eval(feed_dict={ x: mnist.test.images, y_: mnist.test.labels, keep_prob: 1.0})) if __name__ == '__main__': parser = argparse.ArgumentParser() parser.add_argument('--data_dir', type=str, default='/tmp/tensorflow/mnist/input_data', help='Directory for storing input data') FLAGS, unparsed = parser.parse_known_args() tf.app.run(main=main, argv=[sys.argv[0]] + unparsed)

[idata]

Please don't start the tutorial/explanation from tensorflow slim module. Start from reading/loading the check point file.

[idata]

+1

[idata]

AshwinVijayakumar ,

any news regarding this question? I would also very much like an answer.

[idata]

So, what I have to do is train the network and save the graph, then run the movidius compiler that generates a new graph in the ncs format and finally use the API to evaluate an input data ?.

The last post was on November 21, has a new tutorial or documentation of how to train your own network been published? For example MNIST?

[idata]

waiting for this.

[idata]

Also waiting for this workflow for converting a custom-trained tensorflow model into Movidius graph. My understanding is that the example script that is provided in https://movidius.github.io/ncsdk/TensorFlow.html works for the case of creating a model, with an image tensor placeholder input and final output layer that are compatible with movidius. The model is then initialized with variables from a checkpoint and re-saved, so it can be recompiled with mvNCCompile. But when using a model that was already custom-trained from a checkpoint e.g. the model zoo downloads, there does not seem to be a one-to-one mapping of the original model variables, and variable scoping has some extra prefixes that are not correctly mapped if using init_from_checkpoint(). So it would be nice to see a clear example of how to go from a tensorflow frozen graph to a movidius graph. Thanks.

[idata]

Waiting for it too!

[idata]

waiting

[idata]

the only i check into the forum now is to see if there has been an update on this topic

[idata]

+1

[idata]

Hello, I've been working on something similar to this, but changing the gray scale format of MNIST to an RGB format, to use it with the chip. but I get error 34:

[Error 34] Setup Error: Values for input contain placeholder. Pass an absolute value.

What do you mean with "absolute value"?

What is the correct format to declare the input and output tensors?

I searched the Internet for the solution and I can not find it. I've been in this for two weeks. I have not found any developer that has been able to run their own Tensorflow program and share their code. Please upload an example code with a simple program, train the network, save the model and then upload it to the NCS.

It is the minimum of documentation that should exist!!

[idata]

I have been trying to get a Tensorflow ENet segmentation model to work on the NCS, but I keep running into issues such as '[Error 5] Toolkit Error: Stage Details Not Supported: RealDiv'. Is there any comprehensive list of supported tensorflow operations somewhere?

So far, I have had similar errors with the following operations(tf._):

~~_ div (elemwise division => RealDiv)~~

• abs
• log
• pow
• greater
• reciprocal
• gather_nd

Also a weird error where broadcasting did not work and I had to use tf.expand_dims() before multiplying alpha for PReLU layer.

~~The [Error 5] is preventing me from doing division (tf.div() becomes RealDiv), which breaks Spatial Dropout . Adding support for some of the basic math operations listed would be really nice. ~~

I might start a new thread after looking around some more if nobody here has a list which ops are supported. Also, how do we make feature requests? :tongue:

Edit: I updated to version 11, and the div/RealDiv issue seems to have been addressed. I still have not tested other operations, so my question of a list still stands.

[idata]

Are we getting abandoned? So frustrated :/

[idata]

@notbadday @shadySource At the moment, we're working on adding support for these layers and more. Thank you for your patience while we work to add support.

[idata]

I got a solution to this in this thread: https://ncsforum.movidius.com/discussion/536/error-34-values-for-input-contain-placeholder

[idata]

waiting -_-

[idata]

Still waiting. The only thing there is loading Caffe models. Is that really more popular than Tensorflow?

[idata]

@ruze @antoniocappiello There is a document on the NCS doc site that details some general guidelines for compiling a Tensorflow network @ https://movidius.github.io/ncsdk/tf_compile_guidance.html.

[idata]

@ruze, @antoniocappiello, @notbadday, this might serve as a good reference - https://ncsforum.movidius.com/discussion/comment/1579/#Comment_1579

[idata]

@sjbozzo, if you are trying to get MNIST running on NCS, check out this example:

MNIST example: https://github.com/ashwinvijayakumar/ncappzoo/tree/mnist/tensorflow/mnist

[idata]

Is it published @AshwinVijayakumar ?