antorsae

class ScaledWSConv2d(nn.Conv2d):
    """2D Conv layer with Scaled Weight Standardization."""
    def __init__(self, in_channels, out_channels, kernel_size,stride=1, padding=0,dilation=1, groups=1, bias=True, gain=True,eps=1e-4):
        nn.Conv2d.__init__(self, in_channels, out_channels,kernel_size, stride,padding, dilation,groups, bias)
        if gain:
            self.gain = nn.Parameter(torch.ones(self.out_channels, 1, 1, 1))
        else:
            self.gain = None
        # Epsilon, a small constant to avoid dividing by zero.
        self.eps = eps

antorsae

# Keras layer implementation of "Fix your classifier: the marginal value of training the last weight layer"
# by Andres Torrubia, licensed under GPL 3: https://www.gnu.org/licenses/gpl-3.0.en.html
# https://arxiv.org/abs/1801.04540

from keras import backend as K
from keras.engine.topology import Layer
from keras import activations
from keras.initializers import Constant, RandomUniform
import numpy as np
from scipy.linalg import hadamard

antorsae

import os
import sys
import numpy as np
from skimage import io, exposure
import pydicom
import re
import time
from multiprocessing import Pool
from multiprocessing import cpu_count
from tqdm import tqdm

antorsae

# if you have iterm2 for osx (www.iterm2.com) this is a like print(...) for images in the console

import base64
import cStringIO
import numpngw
import numpy as np

def show_image(a):
    if a.dtype != np.uint8:
        a = a.astype(np.uint8)