conv-neural-network

问题 Is the above visualization a rendering of the weights of the first convolutional layer or the activations on a given input image on the first convolutional layer? Below is a visualization of the weights of the first convolutional layer of the Inception v2 model that I've been training for 48 hours: I'm sure I have not converged my model after only 48 hours (on a CPU). Shouldn't those weights begin to smooth out by now, where training accuracy is over 90%? 回答1: According to ImageNet

问题 Is the above visualization a rendering of the weights of the first convolutional layer or the activations on a given input image on the first convolutional layer? Below is a visualization of the weights of the first convolutional layer of the Inception v2 model that I've been training for 48 hours: I'm sure I have not converged my model after only 48 hours (on a CPU). Shouldn't those weights begin to smooth out by now, where training accuracy is over 90%? 回答1: According to ImageNet

问题 While usually people tend to simply resize any image into a square while training a CNN (for example resnet takes a 224x224 square image), that looks ugly to me, especially when the aspect ratio is not around 1. (In fact that might change ground truth eg the label that an expert might give the distorted image could be different than the original one). So now I resize the image to,say, 224x160 , keeping the original ratio, and then I pad the image with 0s (paste it into a random location in a

问题 While usually people tend to simply resize any image into a square while training a CNN (for example resnet takes a 224x224 square image), that looks ugly to me, especially when the aspect ratio is not around 1. (In fact that might change ground truth eg the label that an expert might give the distorted image could be different than the original one). So now I resize the image to,say, 224x160 , keeping the original ratio, and then I pad the image with 0s (paste it into a random location in a

问题 While usually people tend to simply resize any image into a square while training a CNN (for example resnet takes a 224x224 square image), that looks ugly to me, especially when the aspect ratio is not around 1. (In fact that might change ground truth eg the label that an expert might give the distorted image could be different than the original one). So now I resize the image to,say, 224x160 , keeping the original ratio, and then I pad the image with 0s (paste it into a random location in a

问题 While usually people tend to simply resize any image into a square while training a CNN (for example resnet takes a 224x224 square image), that looks ugly to me, especially when the aspect ratio is not around 1. (In fact that might change ground truth eg the label that an expert might give the distorted image could be different than the original one). So now I resize the image to,say, 224x160 , keeping the original ratio, and then I pad the image with 0s (paste it into a random location in a

问题 I am trying to implement a face recognition Siamese Network using the Labelled Faces in the Wild (LFW Dataset in Kaggle). The training data image pairs is stored in the format of : ndarray[ndarray[image1,image2],ndarray[image1,image2]...] and so on. The images are RGB channelled with size of 224*224. There are 2200 training pairs with 1100 match image pairs and 1100 mismatch image pairs. Also, there are 1000 test pairs with 500 match image pairs and 500 mismatch image pairs. I have designed

问题 I am trying to implement a face recognition Siamese Network using the Labelled Faces in the Wild (LFW Dataset in Kaggle). The training data image pairs is stored in the format of : ndarray[ndarray[image1,image2],ndarray[image1,image2]...] and so on. The images are RGB channelled with size of 224*224. There are 2200 training pairs with 1100 match image pairs and 1100 mismatch image pairs. Also, there are 1000 test pairs with 500 match image pairs and 500 mismatch image pairs. I have designed

问题 I have trained a GAN to reproduce CIFAR10 like images. Initially I notice all images cross one batch produced by the generator look always the same, like the picture below: After hours of debugging and comparison to the tutorial which is a great learning source for beginners (https://machinelearningmastery.com/how-to-develop-a-generative-adversarial-network-for-a-cifar-10-small-object-photographs-from-scratch/), I just add only one letter on my original code and the generated images start

问题 I have trained a GAN to reproduce CIFAR10 like images. Initially I notice all images cross one batch produced by the generator look always the same, like the picture below: After hours of debugging and comparison to the tutorial which is a great learning source for beginners (https://machinelearningmastery.com/how-to-develop-a-generative-adversarial-network-for-a-cifar-10-small-object-photographs-from-scratch/), I just add only one letter on my original code and the generated images start