Infopost | 2022.08.03
|
![]() |
Add layer applied to a single-channel image. |
![]() |
Add in RGB. |
![]() |
Add in HSV gives you a hue shift. |
![]() |
Adding the Cb and Cr channels gives this color space even more of a hue shift. |
![]() |
Subtraction layer applied to a single-channel image. |
![]() |
Subtraction in RGB. |
![]() |
Subtraction in HSV. |
![]() |
Subtracting YCbCr is pretty deformative. |
![]() |
Multiply, I guess, makes things darker by the amount of darkness being multiplied (0.0-1.0 values). |
![]() |
RGB multiply looks similar. |
![]() |
In HSV, the multiplication is applied less to brightness and more to saturation. |
![]() |
Likewise YCbCr shifts green. |
![]() |
Average in luminance is pretty straightforward. |
![]() |
Average in RGB also makes sense. |
![]() |
Average in HSV sometimes sees a hue shift. |
![]() |
Average YCbCr works like RGB. |
![]() |
Max in monochrome selects the brighter pixel. |
![]() |
Same in RGB. |
![]() |
It's not as straightforward in HSV where hue and saturation impact which pixel value is used. |
![]() |
Max for YCbCr likewise biases toward purple (red and blue) pixels. |
![]() |
Minimum, of course, selects the darker pixels. |
![]() |
Same with RGB. |
![]() |
In HSV, minimum looks for dark, desaturated pixels with hues happening to be near zero. |
![]() |
YCbCr looks for dark, greenish pixels. |
![]() |
In monochrome images you can see the dark details disappear as pooling selects the brightest pixels. |
![]() |
RGB behaves similar to luminance. |
![]() |
HSV makes the occasional weird selection based on hue and saturation. |
![]() |
Much like with maximum and minimum from the previous section, maxpooling on YCbCr biases toward the purplest pixel. |
![]() |
The jaggies (square artifacts) are less obvious in average pooling. |
![]() |
Edges in RGB look more like antialiasing, flat areas look blurred. |
![]() |
HSV again shows some occasional hue shift. |
![]() |
Like with averaging two images, average pooling a single YCbCr image looks just like RGB. |
![]() |
The ReLu looks pretty close to identical. I may not understand the layer, but expected that each output would be fully connected to the inputs. Hmm. |
![]() |
Sigmoid looks like it inverts the input. |
![]() |
Softplus isn't too fond of the dark parts of the panda. |
![]() |
Tanh seems to have more or less just darkened the input. |
Model: "model" __________________________________________________________________________ Layer (type) Output Shape Param # Connected to ========================================================================== input_1 (InputLayer) [(None, 32, 32, 1)] 0 __________________________________________________________________________ flatten (Flatten) (None, 1024) 0 input_1[0] [0] __________________________________________________________________________ dense (Dense) (None, 1024) 1049600 flatten[0] [0] __________________________________________________________________________ input_2 (InputLayer) [(None, 32, 32, 1)] 0 __________________________________________________________________________ reshape (Reshape) (None, 32, 32, 1) 0 dense[0] [0] ========================================================================== Total params: 1,049,600 Trainable params: 1,049,600 Non-trainable params: 0 __________________________________________________________________________
![]() |
One conv2d layer, kernel size 3, linear activation. |
![]() |
One conv2d layer, kernel size 3, ReLu activation. |
![]() |
One conv2d layer, kernel size 3, sigmoid activation. |
![]() |
One conv2d layer, kernel size 3, softplus activation. |
![]() |
One conv2d layer, kernel size 3, tanh activation. |
![]() |
One conv2d layer, kernel size 7, ReLu activation. |
![]() |
One conv2d layer, kernel size 7, sigmoid activation. |
![]() |
One conv2d layer, kernel size 7, softplus activation. |
![]() |
One conv2d layer, kernel size 7, tanh activation. |
![]() |
Two conv2d layers, kernel size 3, ReLu activation for both. |
![]() |
Two conv2d layers, kernel size 3, ReLu activation and tanh activation. |
![]() |
Two conv2d layers, kernel size 3, tanh activation then ReLu activation. |
![]() |
Two conv2d layers, kernel size 3, tanh activation for both. |
![]() |
Conv2dTranspose, kernel size 2, strides 2, ReLu activation. |
![]() |
Conv2dTranspose, kernel size 2, strides 2, sigmoid activation. |
Conv2dTranspose, kernel size 2, strides 4, tanh activation. |
Conv2dTranspose, kernel size 4, strides 2, ReLu activation. |
Conv2dTranspose, kernel size 4, strides 2, tanh activation. |
Conv2dTranspose, kernel size 8, strides 2, ReLu activation. |
Conv2dTranspose, kernel size 8, strides 2, tanh activation. |
◄ |
2022.07.29
AdhesivesPrepping wood planks for the next project, chasing an elusive Gloomhaven personal quest, concatenate layers in ML, and a little bit of recreation. |
2022.08.14
PopcornWatching this week's political news from the peanut gallery. |
► |
2022.08.19
RewiringPlaying with autoencoders for graphics processing. |
2023.10.08
GranularityLinking internally and externally using text tokens and n-grams. |
2022.01.15
Deep graphicsAn implementation of the Burning Ship fractal and some experiments in creating autoencoders. Changing the style layers in style transfer and combining the outputs into a composite image. Mass Effect 3 Leviathan and Omega DLCs. |
kavita-ganesan.com
A Gentle Introduction to Deep Neural Networks with Python - Kavita Ganesan, PhDThis article examines the parts that make up neural networks and deep neural networks, as well as the fundamental different types of models (e.g. regression), their constituent parts (and how they contribute to model accuracy), and which tasks they are designed to learn. |
www.paepper.com
Pyramidal Convolution: Rethinking Convolutional Neural Networks for Visual Recognition :: Päpper's Machine Learning Blog - Th...Today's paper: Pyramidal Convolution by Duta et al. This is the third paper of the new series Deep Learning Papers visualized and it's about using convolutions in a pyramidal style to capture information of different magnifications from an image. The authors show how a pyramidal convolution can be constructed and apply it to several problems in the visual domain. What's really interesting is that the number of parameters can be kept the same while performance tends to improve. |
pyimagesearch.com
Keras and Convolutional Neural Networks (CNNs) - PyImageSearchThis gentle guide will show you how to implement, train, and evaluate your first Convolutional Neural Network (CNN) with Keras and deep learning. |