Delete autoregressor_app.py
Browse files- autoregressor_app.py +0 -195
autoregressor_app.py
DELETED
@@ -1,195 +0,0 @@
|
|
1 |
-
import streamlit as st
|
2 |
-
|
3 |
-
import numpy as np
|
4 |
-
import matplotlib.pyplot as plt
|
5 |
-
import pandas as pd
|
6 |
-
import seaborn as sns
|
7 |
-
import warnings
|
8 |
-
warnings.filterwarnings('ignore')
|
9 |
-
# %matplotlib inline
|
10 |
-
|
11 |
-
import tensorflow
|
12 |
-
print (tensorflow.__version__)
|
13 |
-
|
14 |
-
st.header("Welcome to the Generative Playground")
|
15 |
-
st.write("This is an autoregressor model on cifar10 data set, with 50 epochs and 16 batch size trained only. RTX GPU is used to train the model.")
|
16 |
-
|
17 |
-
from tensorflow.keras.datasets import mnist,cifar10
|
18 |
-
|
19 |
-
|
20 |
-
(trainX, trainy), (testX, testy) = cifar10.load_data()
|
21 |
-
|
22 |
-
print('Training data shapes: X=%s, y=%s' % (trainX.shape, trainy.shape))
|
23 |
-
print('Testing data shapes: X=%s, y=%s' % (testX.shape, testy.shape))
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
for k in range(4):
|
28 |
-
fig = plt.figure(figsize=(9,6))
|
29 |
-
for j in range(9):
|
30 |
-
i = np.random.randint(0, 10000)
|
31 |
-
plt.subplot(990 + 1 + j)
|
32 |
-
plt.imshow(trainX[i], cmap='gray_r')
|
33 |
-
# st.pyplot(fig)
|
34 |
-
plt.axis('off')
|
35 |
-
#plt.title(trainy[i])
|
36 |
-
plt.show()
|
37 |
-
st.pyplot(fig)
|
38 |
-
|
39 |
-
|
40 |
-
# asdfaf
|
41 |
-
|
42 |
-
trainX = np.where(trainX < (0.33 * 256), 0, 1)
|
43 |
-
train_data = trainX.astype(np.float32)
|
44 |
-
|
45 |
-
testX = np.where(testX < (0.33 * 256), 0, 1)
|
46 |
-
test_data = testX.astype(np.float32)
|
47 |
-
|
48 |
-
train_data = np.reshape(train_data, (50000, 32, 32, 3))
|
49 |
-
test_data = np.reshape(test_data, (10000, 32, 32, 3))
|
50 |
-
|
51 |
-
print (train_data.shape, test_data.shape)
|
52 |
-
|
53 |
-
|
54 |
-
import tensorflow
|
55 |
-
|
56 |
-
class PixelConvLayer(tensorflow.keras.layers.Layer):
|
57 |
-
def __init__(self, mask_type, **kwargs):
|
58 |
-
super(PixelConvLayer, self).__init__()
|
59 |
-
self.mask_type = mask_type
|
60 |
-
self.conv = tensorflow.keras.layers.Conv2D(**kwargs)
|
61 |
-
|
62 |
-
def build(self, input_shape):
|
63 |
-
# Build the conv2d layer to initialize kernel variables
|
64 |
-
self.conv.build(input_shape)
|
65 |
-
# Use the initialized kernel to create the mask
|
66 |
-
kernel_shape = self.conv.kernel.get_shape()
|
67 |
-
self.mask = np.zeros(shape=kernel_shape)
|
68 |
-
self.mask[: kernel_shape[0] // 2, ...] = 1.0
|
69 |
-
self.mask[kernel_shape[0] // 2, : kernel_shape[1] // 2, ...] = 1.0
|
70 |
-
if self.mask_type == "B":
|
71 |
-
self.mask[kernel_shape[0] // 2, kernel_shape[1] // 2, ...] = 1.0
|
72 |
-
|
73 |
-
def call(self, inputs):
|
74 |
-
self.conv.kernel.assign(self.conv.kernel * self.mask)
|
75 |
-
return self.conv(inputs)
|
76 |
-
|
77 |
-
|
78 |
-
# Next, we build our residual block layer.
|
79 |
-
# This is just a normal residual block, but based on the PixelConvLayer.
|
80 |
-
class ResidualBlock(tensorflow.keras.layers.Layer):
|
81 |
-
def __init__(self, filters, **kwargs):
|
82 |
-
super(ResidualBlock, self).__init__(**kwargs)
|
83 |
-
self.conv1 = tensorflow.keras.layers.Conv2D(
|
84 |
-
filters=filters, kernel_size=1, activation="relu"
|
85 |
-
)
|
86 |
-
self.pixel_conv = PixelConvLayer(
|
87 |
-
mask_type="B",
|
88 |
-
filters=filters // 2,
|
89 |
-
kernel_size=3,
|
90 |
-
activation="relu",
|
91 |
-
padding="same",
|
92 |
-
)
|
93 |
-
self.conv2 = tensorflow.keras.layers.Conv2D(
|
94 |
-
filters=filters, kernel_size=1, activation="relu"
|
95 |
-
)
|
96 |
-
|
97 |
-
def call(self, inputs):
|
98 |
-
x = self.conv1(inputs)
|
99 |
-
x = self.pixel_conv(x)
|
100 |
-
x = self.conv2(x)
|
101 |
-
return tensorflow.keras.layers.add([inputs, x])
|
102 |
-
|
103 |
-
inputs = tensorflow.keras.Input(shape=(32,32,3))
|
104 |
-
x = PixelConvLayer(
|
105 |
-
mask_type="A", filters=128, kernel_size=7, activation="relu", padding="same"
|
106 |
-
)(inputs)
|
107 |
-
|
108 |
-
for _ in range(5):
|
109 |
-
x = ResidualBlock(filters=128)(x)
|
110 |
-
|
111 |
-
for _ in range(2):
|
112 |
-
x = PixelConvLayer(
|
113 |
-
mask_type="B",
|
114 |
-
filters=128,
|
115 |
-
kernel_size=1,
|
116 |
-
strides=1,
|
117 |
-
activation="relu",
|
118 |
-
padding="valid",
|
119 |
-
)(x)
|
120 |
-
|
121 |
-
out = tensorflow.keras.layers.Conv2D(
|
122 |
-
filters=3, kernel_size=1, strides=1, activation="sigmoid", padding="valid"
|
123 |
-
)(x)
|
124 |
-
|
125 |
-
pixel_cnn = tensorflow.keras.Model(inputs, out)
|
126 |
-
pixel_cnn.summary()
|
127 |
-
|
128 |
-
adam = tensorflow.keras.optimizers.Adam(learning_rate=0.0005)
|
129 |
-
pixel_cnn.compile(optimizer=adam, loss="binary_crossentropy")
|
130 |
-
|
131 |
-
|
132 |
-
# %%
|
133 |
-
import os
|
134 |
-
checkpoint_path = "training_1/cp.ckpt"
|
135 |
-
checkpoint_dir = os.path.dirname(checkpoint_path)
|
136 |
-
|
137 |
-
|
138 |
-
pixel_cnn.load_weights(checkpoint_path)
|
139 |
-
|
140 |
-
|
141 |
-
# %% [markdown]
|
142 |
-
# # Display Results 81 images
|
143 |
-
|
144 |
-
# %%
|
145 |
-
#from IPython.display import Image, display
|
146 |
-
from tqdm import tqdm_notebook
|
147 |
-
|
148 |
-
|
149 |
-
# Create an empty array of pixels.
|
150 |
-
batch = 81
|
151 |
-
pixels = np.zeros(shape=(batch,) + (pixel_cnn.input_shape)[1:])
|
152 |
-
batch, rows, cols, channels = pixels.shape
|
153 |
-
|
154 |
-
print(pixels.shape)
|
155 |
-
|
156 |
-
|
157 |
-
import time
|
158 |
-
|
159 |
-
# progress_text = "Operation in progress. Please wait."
|
160 |
-
# my_bar = st.progress(0, progress_text)
|
161 |
-
st.caption("Generating..... pls.. wait.. :)")
|
162 |
-
my_bar = st.progress(0)
|
163 |
-
|
164 |
-
|
165 |
-
# Iterate over the pixels because generation has to be done sequentially pixel by pixel.
|
166 |
-
for row in range(rows):
|
167 |
-
for col in range(cols):
|
168 |
-
for channel in range(channels):
|
169 |
-
time.sleep(0.01)
|
170 |
-
# Feed the whole array and retrieving the pixel value probabilities for the next
|
171 |
-
# pixel.
|
172 |
-
probs = pixel_cnn.predict(pixels)[:, row, col, channel]
|
173 |
-
# Use the probabilities to pick pixel values and append the values to the image
|
174 |
-
# frame.
|
175 |
-
pixels[:, row, col, channel] = tensorflow.math.ceil(
|
176 |
-
probs - tensorflow.random.uniform(probs.shape)
|
177 |
-
)
|
178 |
-
my_bar.progress(row+1)
|
179 |
-
time.sleep(1)
|
180 |
-
|
181 |
-
counter = 0
|
182 |
-
for i in range(4):
|
183 |
-
figout = plt.figure(figsize=(9,6))
|
184 |
-
for j in range(9):
|
185 |
-
plt.subplot(990 + 1 + j)
|
186 |
-
plt.imshow(pixels[counter,:,:,0])#, cmap='gray_r')
|
187 |
-
counter += 1
|
188 |
-
plt.axis('off')
|
189 |
-
plt.show()
|
190 |
-
st.pyplot(figout)
|
191 |
-
|
192 |
-
# %%
|
193 |
-
|
194 |
-
|
195 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|