conv-neural-network

问题 I am currently looking for a way i can combine the output of multiple model into one model, I need to create a CNN network that does classification. The image is separated into sections (as seen by the colors), each section is given as input to a certain model (1,2,3,4) the structure of each model is the same, but each section is given to a separate model to ensure that the the same weight is not applied on whole image - My attempt to avoid full weight sharing, and keeping the weight sharing

问题 I am trying to code and learn different neural network models. I am having a lot of complication with input dimensionality. I am looking for some tutorial that shows differences in layers and how to set input and outputs for each layers. 回答1: Keras documentation shows you all the input_shape s expected by each layer. In Keras, you'll see input shapes in these forms: input_shape defined by user in layers shapes shown in summaries and others array shapes tensor shapes Input shape defined by

问题 I am new to Keras and I am trying to get the weights in Keras. I know how to do it in Tensorflow in Python. Code: data = np.array(attributes, 'int64') target = np.array(labels, 'int64') feature_columns = [tf.contrib.layers.real_valued_column("", dimension=2, dtype=tf.float32)] learningRate = 0.1 epoch = 10000 # https://www.tensorflow.org/api_docs/python/tf/metrics validation_metrics = { "accuracy": tf.contrib.learn.MetricSpec(metric_fn = tf.contrib.metrics.streaming_accuracy , prediction_key

问题 I have a little comprehension problem with CNN. And I'm not quite sure how many filters and thus weights are trained. Example: I have an input layer with the 32x32 pixels and 3 channels (i.e. shape of (32,32,3) ). Now I use a 2D-convolution layer with 10 filters of shape (4,4) . So I end up with 10 channels each with shape of (28,28) , but do I now train a separate filter for each input channel or are they shared? Do I train 3x10x4x4 weights or do I train 10x4x4 weights? 回答1: You can find out

问题 I have built a Convolutional LSTM model using Tensorflow ConvLSTMCell(), tf.nn.dynamic_rnn(), and tf.contrib.legacy_seq2seq.rnn_decoder(). I have 3 layers of encoder, and 3 layers of decoder, the initial states of decoders come from the final states of encoders. I have 128, 64, and 64 filters for layer 1, layer 2, and layer 3 respectively. finally, I concatenate the outputs of decoders and pass them through a convolution layer to decrease the number of channels to one. and then I apply the

问题 I need some help to create a CaffeDB for siamese CNN out of a plain directory with images and label-text-file. Best would be a python-way to do it. The problem is not to walk through the directory and making pairs of images. My problem is more of making a CaffeDB out of those pairs. So far I only used convert_imageset to create a CaffeDB out of an image directory. Thanks for help! 回答1: Why don't you simply make two datasets using good old convert_imagest ? layer { name: "data_a" top: "data_a"

问题 This is a classification model for ten categories of pictures. My code has three files, one is the CNN model convNet.py, one is read_TFRecord.py to read data, one is train.py to train and evaluation model. Training set of samples of 80,000, validation set of sample of 20,000. Question: In the first epoch: training loss = 2.11, train accuracy = 25.61% validation loss = 3.05, validation accuracy = 8.29% Why validation loss are significantly different right from the start? And why the validation

问题 I am trying to train CNN (Sklearn Neural Network). I am having 4 images of 128 x 128 pixels. shape -> (4, 128, 128) I am reading images like - in1 = misc.imread('../data/Train_Data/train-1.jpg', mode='L', flatten=True)/255. in2 = misc.imread('../data/Train_Data/train-2.jpg', mode='L', flatten=True)/255. in3 = misc.imread('../data/Train_Data/train-3.jpg', mode='L', flatten=True)/255. in4 = misc.imread('../data/Train_Data/train-4.jpg', mode='L', flatten=True)/255. Then numpy array is created

问题 I have a really weird problem I want to explain to you. I am not sure if this is a topic for SO but I hope it will be in the end. My general problem task is depth estimation, i.e. I have an image as input and its corresponding ground_truth (depth image). Then I have my net (which should be considered as black box) and my last layers. First of all depth estimation is rather a regression task than a classification task. Therefore I decided to use a EuclideanLoss layer where my num_output of my

问题 I'm trying to do depth estimation with CNNs (this is my ultimate goal), but a problem that i found is: I just did image classifications with CNNs, using for example "CIFAR-10", "MNIST", "Cats vs Dogs", etc. To do depth estimation I need to output a new image (the NYUv2 dataset has the labeled images). So, I'll input an image like 256x256x3 and need to output another image with for example 228x228x3. What I need to do? Can I just do the convolutions for a while and after that decrease the