File size: 4,699 Bytes
7a3f743 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 |
# =============================================================================
# Project: WHITE-BOX-CARTOONIZATION
# Authors: Amey Thakur & Mega Satish
# Date: 2021-08-28
# Repository: https://github.com/Amey-Thakur/WHITE-BOX-CARTOONIZATION
# Profiles: https://github.com/Amey-Thakur | https://github.com/msatmod
# =============================================================================
"""
network.py
=============================================================================
This file defines the Neural Network architecture used for cartoonization.
It uses a "U-Net" based Generator with Residual Blocks.
Key Components:
1. Convolutional Layers: To extract features (edges, textures) from the image.
2. Leaky ReLU: Activation function to introduce non-linearity.
3. Residual Blocks (ResBlock): To help the network learn complex transformations without losing original details.
4. U-Net Structure: Downsamples the image to understand global context, then upsamples it back to original size.
=============================================================================
"""
import tensorflow as tf
import numpy as np
try:
import tf_slim as slim
except ImportError:
try:
import tensorflow.contrib.slim as slim
except ImportError:
print("Error: Could not import slim. Please install tf-slim.")
def resblock(inputs, out_channel=32, name='resblock'):
"""
Defines a Residual Block.
Input -> [Conv -> ReLU -> Conv] + Input -> Output
This "skip connection" (+ Input) prevents the gradient from vanishing
and allows the network to learn "residuals" (changes) rather than constructing from scratch.
"""
with tf.compat.v1.variable_scope(name):
# First Convolution
x = slim.convolution2d(inputs, out_channel, [3, 3],
activation_fn=None, scope='conv1')
x = tf.nn.leaky_relu(x)
# Second Convolution
x = slim.convolution2d(x, out_channel, [3, 3],
activation_fn=None, scope='conv2')
# Add the original input back to the result (Skip Connection)
return x + inputs
def unet_generator(inputs, channel=32, num_blocks=4, name='generator', reuse=False):
"""
Defines the Generator Network.
Structure: Encoder -> Bottleneck (ResBlocks) -> Decoder
"""
with tf.compat.v1.variable_scope(name, reuse=reuse):
# --- ENCODER (Downsampling) ---
# Reduce the spatial size (Height/Width) but increase the depth (Channels)
# Initial Convolution (7x7 kernel to capture large features)
x0 = slim.convolution2d(inputs, channel, [7, 7], activation_fn=None)
x0 = tf.nn.leaky_relu(x0)
# Downsample 1
x1 = slim.convolution2d(x0, channel, [3, 3], stride=2, activation_fn=None)
x1 = tf.nn.leaky_relu(x1)
x1 = slim.convolution2d(x1, channel*2, [3, 3], activation_fn=None)
x1 = tf.nn.leaky_relu(x1)
# Downsample 2
x2 = slim.convolution2d(x1, channel*2, [3, 3], stride=2, activation_fn=None)
x2 = tf.nn.leaky_relu(x2)
x2 = slim.convolution2d(x2, channel*4, [3, 3], activation_fn=None)
x2 = tf.nn.leaky_relu(x2)
# --- BOTTLENECK (Processing) ---
# Apply multiple Residual Blocks to process the image features (the "Cartoonizing" logic)
for idx in range(num_blocks):
x2 = resblock(x2, out_channel=channel*4, name='block_{}'.format(idx))
# --- DECODER (Upsampling) ---
# Increase spatial size back to original resolution
x2 = slim.convolution2d(x2, channel*2, [3, 3], activation_fn=None)
x2 = tf.nn.leaky_relu(x2)
# Upsample 1
h1, w1 = tf.shape(x2)[1], tf.shape(x2)[2]
x3 = tf.compat.v1.image.resize_bilinear(x2, (h1*2, w1*2)) # Double the size
x3 = slim.convolution2d(x3+x1, channel*2, [3, 3], activation_fn=None) # +x1 is a Skip Connection from Encoder
x3 = tf.nn.leaky_relu(x3)
x3 = slim.convolution2d(x3, channel, [3, 3], activation_fn=None)
x3 = tf.nn.leaky_relu(x3)
# Upsample 2
h2, w2 = tf.shape(x3)[1], tf.shape(x3)[2]
x4 = tf.compat.v1.image.resize_bilinear(x3, (h2*2, w2*2)) # Double the size again
x4 = slim.convolution2d(x4+x0, channel, [3, 3], activation_fn=None) # +x0 is a Skip Connection from Input
x4 = tf.nn.leaky_relu(x4)
# Final Convolution to produce RGB image (3 channels)
x4 = slim.convolution2d(x4, 3, [7, 7], activation_fn=None)
return x4
if __name__ == '__main__':
pass
|