Mirrors_Framework/FSRCNN-TensorFlow
1
0
Fork 0
mirror of https://github.com/igv/FSRCNN-TensorFlow.git synced 2026-08-18 00:57:28 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Layer for predicting local sub-band residuals

Browse Source
This commit is contained in:
igv
2018-05-12 12:27:10 +03:00
parent 5274a73ebe
commit 1b35ecb2cd
2 changed files with 71 additions and 40 deletions
Download Patch File Download Diff File Expand all files Collapse all files
FSRCNN.py
+11 -6
View File
@@ -51,22 +51,27 @@ class Model(object):
conv = tf.nn.conv2d(conv, weights, strides=[1,1,1,1], padding='SAME', data_format='NHWC')
conv = tf.nn.bias_add(conv, biases, data_format='NHWC')
if i == m + 2:
conv = self.prelu(conv, m + 3)
weights = tf.get_variable('w{}'.format(m + 3), shape=[1, 1, s, s], initializer=tf.variance_scaling_initializer(2))
biases = tf.get_variable('b{}'.format(m + 3), initializer=tf.zeros([s]))
conv = tf.nn.conv2d(conv, weights, strides=[1,1,1,1], padding='SAME', data_format='NHWC')
conv = tf.nn.bias_add(conv, biases, data_format='NHWC')
conv = tf.add(conv, features)
scope.reuse_variables()
conv = self.prelu(conv, 2)
# Expanding
if self.model_params[1] > 0:
expand_weights = tf.get_variable('w{}'.format(m + 3), shape=[1, 1, s, d], initializer=tf.variance_scaling_initializer(2))
expand_biases = tf.get_variable('b{}'.format(m + 3), initializer=tf.zeros([d]))
expand_weights = tf.get_variable('w{}'.format(m + 4), shape=[1, 1, s, d], initializer=tf.variance_scaling_initializer(2))
expand_biases = tf.get_variable('b{}'.format(m + 4), initializer=tf.zeros([d]))
conv = tf.nn.conv2d(conv, expand_weights, strides=[1,1,1,1], padding='SAME', data_format='NHWC')
conv = tf.nn.bias_add(conv, expand_biases, data_format='NHWC')
conv = self.prelu(conv, m + 3)
conv = self.prelu(conv, m + 4)
# Deconvolution
deconv_size = self.radius * self.scale * 2 + 1
deconv_weights = tf.get_variable('w{}'.format(m + 4), shape=[deconv_size, deconv_size, 1, d], initializer=tf.variance_scaling_initializer(0.01))
deconv_biases = tf.get_variable('b{}'.format(m + 4), initializer=tf.zeros([1]))
deconv_weights = tf.get_variable('w{}'.format(m + 5), shape=[deconv_size, deconv_size, 1, d], initializer=tf.variance_scaling_initializer(0.01))
deconv_biases = tf.get_variable('b{}'.format(m + 5), initializer=tf.zeros([1]))
deconv_output = [self.batch, self.label_size, self.label_size, self.c_dim]
deconv_stride = [1, self.scale, self.scale, 1]
deconv = tf.nn.conv2d_transpose(conv, deconv_weights, output_shape=deconv_output, strides=deconv_stride, padding='SAME', data_format='NHWC')
@@ -75,7 +80,7 @@ class Model(object):
if self.GRL:
# Deconvolution 2
upsample_filter = bilinear_upsample_weights(self.scale, self.c_dim)
self.biases['b{}'.format(m + 5)] = tf.get_variable('b{}'.format(m + 5), initializer=tf.constant(1, shape=[1]))
self.biases['b{}'.format(m + 6)] = tf.get_variable('b{}'.format(m + 6), initializer=tf.constant(1, shape=[1]))
deconv_output = [self.batch, self.label_size, self.label_size, self.c_dim]
deconv_stride = [1, self.scale, self.scale, 1]
img = tf.image.resize_image_with_crop_or_pad(self.images, self.image_size, self.image_size)
gen.py
+60 -34
View File
@@ -53,25 +53,32 @@ def header3(file, r, mi, m, n, s, inp):
base_header(file)
file.write('//!DESC mapping {}_{}\n'.format(mi + 1, (n//4)%(s//4) + 1))
for i in range(s//4):
file.write('//!BIND {}{}\n'.format(inp if r == 0 and mi == 0 else "MODEL", i+1 + (0 if (r * m + mi) % 2 == 0 else 20)))
if mi == m-1 and (mi+1)*(r+1) > 1:
file.write('//!BIND {}{}\n'.format(inp, (n//4)%(s//4) + 1))
file.write('//!BIND {}{}\n'.format(inp, i+1 + (0 if (r * m + mi) % 2 == 0 else 20)))
file.write('//!SAVE MODEL{}\n'.format((n//4)%(s//4) + 1 + (20 if (r * m + mi) % 2 == 0 else 0)))
file.write('//!COMPONENTS 4\n')
def header3_1(file, r, mi, m, n, s, inp):
base_header(file)
file.write('//!DESC sub-band residuals {}\n'.format((n//4)%(s//4) + 1))
for i in range(s//4):
file.write('//!BIND MODEL{}\n'.format(i + 1 + (20 if (r * m + mi) % 2 == 0 else 0)))
file.write('//!BIND {}{}\n'.format(inp, (n//4)%(s//4) + 1))
file.write('//!SAVE RES{}\n'.format((n//4)%(s//4) + 1))
file.write('//!COMPONENTS 4\n')
def header4(file, s, m, r, n, d):
base_header(file)
file.write('//!DESC expanding {}\n'.format((n//4)%(d//4) + 1))
for i in range(s//4):
file.write('//!BIND MODEL{}\n'.format(i + 1 + (20 if (r * m) % 2 == 1 else 0)))
file.write('//!SAVE EXPANDED{}\n'.format((n//4)%(d//4) + 1 + (20 if (r * m) % 2 == 1 else 0)))
file.write('//!BIND RES{}\n'.format(i + 1))
file.write('//!SAVE EXPANDED{}\n'.format((n//4)%(d//4) + 1))
file.write('//!COMPONENTS 4\n')
def header5(file, m, r, n, d, inp):
base_header(file)
file.write('//!DESC sub-pixel convolution {}\n'.format((n//4)%(d//4) + 1))
file.write('//!BIND {}{}\n'.format(inp, (n//4)%(d//4) + 1 + (20 if (r * m) % 2 == 1 else 0)))
file.write('//!SAVE {}{}\n'.format(inp, (n//4)%(d//4) + 1 + (20 if (r * m) % 2 == 1 else 0)))
file.write('//!BIND {}{}\n'.format(inp, (n//4)%(d//4) + 1))
file.write('//!SAVE {}{}\n'.format(inp, (n//4)%(d//4) + 1))
file.write('//!COMPONENTS 4\n')
def header6(file, m, r, d, inp, grl):
@@ -82,7 +89,7 @@ def header6(file, m, r, d, inp, grl):
if grl:
file.write('//!BIND HOOKED\n')
for i in range(d//4):
file.write('//!BIND {}{}\n'.format(inp, i+1 + (20 if (r * m) % 2 == 1 else 0)))
file.write('//!BIND {}{}\n'.format(inp, i + 1))
file.write('//!OFFSET -{}.0 -{}.0\n'.format(scale//2, scale//2))
def main():
@@ -137,15 +144,16 @@ def main():
file.write('}\n\n')
# Mapping layers
inp = "SHRINKED" if shrinking else "FEATURE"
for ri in range(r):
for mi in range(m):
tex_name = inp if ri == 0 and mi == 0 else "RES" if ri > 0 and mi == 0 else "MODEL"
ln = get_line_number("w{}".format(mi + 3), fname)
weights = read_weights(fname, ln, s*9)
ln = get_line_number("b{}".format(mi + 3), fname)
biases = read_weights(fname, ln)
inp = "SHRINKED" if shrinking else "FEATURE"
for n in range(0, s, 4):
header3(file, ri, mi, m, n, s, inp)
header3(file, ri, mi, m, n, s, tex_name)
file.write('vec4 hook()\n')
file.write('{\n')
file.write('vec4 res = vec4({});\n'.format(format_weights(biases[0], n)))
@@ -158,26 +166,43 @@ def main():
file.write('res += mat4({},{},{},{}) * {}{}_texOff(vec2({},{}));\n'.format(
format_weights(weights[l], n), format_weights(weights[l+1], n),
format_weights(weights[l+2], n), format_weights(weights[l+3], n),
inp if ri == 0 and mi == 0 else "MODEL",
idx + 1 + (20 if (ri * m + mi) % 2 == 1 else 0), x, y))
ln = get_line_number("alpha{}".format(3 if mi == m - 1 else mi + 4), fname)
tex_name, idx + 1 + (20 if (ri * m + mi) % 2 == 1 else 0), x, y))
ln = get_line_number("alpha{}".format(m + 3 if mi == m - 1 else mi + 4), fname)
alphas = read_weights(fname, ln)
if mi == m - 1:
file.write('res += {}{}_texOff(0);\n'.format(inp, (n//4)%(s//4) + 1))
if ri == r - 1:
ln = get_line_number("alpha2", fname)
alphas = read_weights(fname, ln)
file.write('res = max(res, vec4(0.0)) + vec4({}) * min(res, vec4(0.0));\n'.format(format_weights(alphas[0], n)))
file.write('return res;\n')
file.write('}\n\n')
if mi == m - 1:
ln = get_line_number("w{}".format(m + 3), fname)
weights = read_weights(fname, ln, s*(mi+2))
ln = get_line_number("b{}".format(m + 3), fname)
biases = read_weights(fname, ln)
for n in range(0, s, 4):
header3_1(file, ri, mi, m, n, s, inp)
file.write('vec4 hook()\n')
file.write('{\n')
file.write('vec4 res = vec4({});\n'.format(format_weights(biases[0], n)))
for l in range(0, s, 4):
file.write('res += mat4({},{},{},{}) * MODEL{}_texOff(0);\n'.format(
format_weights(weights[l], n), format_weights(weights[l+1], n),
format_weights(weights[l+2], n), format_weights(weights[l+3], n),
l//4 + 1 + (20 if (ri * m + mi) % 2 == 0 else 0)))
file.write('res += {}{}_texOff(0);\n'.format(inp, (n//4)%(s//4) + 1))
if ri == r - 1:
ln = get_line_number("alpha2", fname)
alphas = read_weights(fname, ln)
file.write('res = max(res, vec4(0.0)) + vec4({}) * min(res, vec4(0.0));\n'.format(format_weights(alphas[0], n)))
file.write('return res;\n')
file.write('}\n\n')
if shrinking:
# Expanding layer
ln = get_line_number("w{}".format(m + 3), fname)
ln = get_line_number("w{}".format(m + 4), fname)
weights = read_weights(fname, ln, d)
ln = get_line_number("b{}".format(m + 3), fname)
ln = get_line_number("b{}".format(m + 4), fname)
biases = read_weights(fname, ln)
ln = get_line_number("alpha{}".format(m + 3), fname)
ln = get_line_number("alpha{}".format(m + 4), fname)
alphas = read_weights(fname, ln)
for n in range(0, d, 4):
header4(file, s, m, r, n, d)
@@ -185,14 +210,14 @@ def main():
file.write('{\n')
file.write('vec4 res = vec4({});\n'.format(format_weights(biases[0], n)))
for l in range(0, s, 4):
file.write('res += mat4({},{},{},{}) * MODEL{}_texOff(vec2(0.0));\n'.format(format_weights(weights[l], n), format_weights(weights[l+1], n), format_weights(weights[l+2], n), format_weights(weights[l+3], n),
l//4 + 1 + (20 if (r * m) % 2 == 1 else 0)))
file.write('res += mat4({},{},{},{}) * RES{}_texOff(vec2(0.0));\n'.format(format_weights(weights[l], n), format_weights(weights[l+1], n), format_weights(weights[l+2], n), format_weights(weights[l+3], n),
l//4 + 1))
file.write('res = max(res, vec4(0.0)) + vec4({}) * min(res, vec4(0.0));\n'.format(format_weights(alphas[0], n)))
file.write('return res;\n')
file.write('}\n\n')
# Sub-pixel convolution
ln = get_line_number("w{}".format(m + 4), fname)
ln = get_line_number("w{}".format(m + 5), fname)
weights = read_weights(fname, ln, dsize**2)
x=list(reversed(range(scale)))
@@ -212,7 +237,7 @@ def main():
weights = list(reversed(weights))
sort = [weights[id[l]].strip(",") for l in range(0, len(id))]
inp = "EXPANDED" if shrinking else "MODEL"
inp = "EXPANDED" if shrinking else "RES"
for n in range(0, d, 4):
header5(file, m, r, n, d, inp)
file.write('vec4 hook()\n')
@@ -228,26 +253,27 @@ def main():
for xi, x in enumerate(range(-radius + (0 if j == 0 and dsize % 2 == 1 else 1), radius + 1)):
l = yi * s1 + xi
file.write('dot(vec4({}), {}{}_texOff(vec2({},{}))){}\n'.format(format_weights(sort[l+total], n), inp,
(n//4)%(d//4) + 1 + (20 if (r * m) % 2 == 1 else 0), x, y, ';' if l == s1 * s2 - 1 else '+'))
(n//4)%(d//4) + 1, x, y, ';' if l == s1 * s2 - 1 else '+'))
total = total + l + 1
file.write('return res;\n')
file.write('}\n\n')
# Aggregation
ln = get_line_number("b{}".format(m + 4), fname)
ln = get_line_number("b{}".format(m + 5), fname)
biases = read_weights(fname, ln)
grl = get_line_number("b{}".format(m + 5), fname)
grl = get_line_number("b{}".format(m + 6), fname)
header6(file, m, r, d, inp, grl)
file.write('vec4 hook()\n')
file.write('{\n')
file.write('float res = {};\n'.format(float(biases[0])))
v = 1 + (20 if (r * m) % 2 == 1 else 0)
file.write('vec2 fcoord = fract({}{}_pos * {}{}_size);\n'.format(inp, v, inp, v))
file.write('vec2 base = {}{}_pos + (vec2(0.5) - fcoord) * {}{}_pt;\n'.format(inp, v, inp, v))
file.write('vec2 fcoord = fract({}1_pos * {}1_size);\n'.format(inp, inp))
file.write('vec2 base = {}1_pos + (vec2(0.5) - fcoord) * {}1_pt;\n'.format(inp, inp))
file.write('ivec2 index = ivec2(fcoord * vec2({}));\n'.format(scale))
file.write('res += ({}{}_tex(base)'.format(inp, v))
for i in range(d//4-1):
file.write('+{}{}_tex(base)'.format(inp, i + 1 + v))
file.write('res += (')
for i in range(d//4):
if i > 0:
file.write('+')
file.write('{}{}_tex(base)'.format(inp, i + 1))
file.write(')[index.y * {} + index.x];\n'.format(scale))
if grl: