fred-003 Goto Github PK

creating-the-image-features

def read_image(src): img = cv2.imread(src) if img is None: raise FileNotFoundError img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) return img

home-work

how-to-improve-gan-technology

How to improve GAN technology

image

improved-techniques-for-training-gan-posted-in-generative-dog-images-a-year-ago-arrow_drop_up

1. Normalize the input: normalize the images between -1 and 1 Tanh as the last layer of the generator output 2. Batchnorm Construct different mini-batches for real and fake, i.e. each mini-batch needs to contain only all real images or all generated images. when batchnorm is not an option use instance normalization (for each sample, subtract mean and divide by standard deviation). 3. Avoid Sparse Gradients the stability of the GAN game suffers if you have sparse gradients LeakyReLU = good (in both G and D) For Downsampling, use: Average Pooling, Conv2d + stride 4: Use the ADAM Optimizer Perhaps the real papers might have a different opinion, this literally makes a remarkable improvement Use SGD for discriminator and ADAM for generator. To decrease instability issues decrease the learning rate to 0.0002 (from 0.001) and the momentum/beta1 to 0.5 (from 0.9) for Adam. Apart from these, I also came across a few blogs that would also help:

python-project-structure-and-dump-into-a-valid-stand-alone-script

the-public-private-mifid-is-demonstrated-below

use-your-training-skills-to-create-images-rather-than-identify-them

Use your training skills to create images, rather than identify them

you-ll-enable-us-to-create-more-experiments-across-a-variety-of-domains-in-the-future.

you’ll enable us to create more experiments across a variety of domains in the future.