net.get_model().summary()
img = cv2.imread('div2k_test.png')
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
r = DATA.patch_size - img.shape[0] % DATA.patch_size
c = DATA.patch_size - img.shape[1] % DATA.patch_size
img = np.pad(img, [(0, r), (0, c), (0, 0)], 'constant')
Image.fromarray(img).save("div2k_padded_test.png")
lr_img = cv2.resize(img, (int(img.shape[1] / 2), int(img.shape[0] / 2)),
cv2.INTER_CUBIC)
Image.fromarray(lr_img).save("div2k_padded_test_lr.png")
p, r, c = DATA.patchify(lr_img, scale=2)
for i in range(chk + 1, tryout):
print("tryout no: ", i)
samplev = np.random.random_integers(0,
DATA.training_patches_2x.shape[0] - 1,
sample)
net.fit(DATA.training_patches_2x[samplev],
DATA.training_patches_Y[samplev], batch_size, epochs)
net.get_model().save_weights('model_iter' + str(i) + '.h5')
g = net.get_model().predict(np.array(p))
gen = DATA.reconstruct(g, r, c, scale=1)
gen[gen > 255] = 255
gen[gen < 0] = 0
Image.fromarray(gen).save("main_div2k_gen_" + str(i) + ".jpeg")
print("Reconstruction Gain:", PSNRLossnp(img, gen))
net.visualize()
net.get_model().summary()
# Temp stuff for barcode #####
os.mkdir('bargen')
loc = '/home/sanchit/Barcodes/HR/detected_barcodes'
for f in os.listdir(loc):
if not os.path.isdir(f):
filename = loc + '/' + f
img = cv2.imread(filename)
r = 32 - img.shape[0] % 32
c = 32 - img.shape[1] % 32
img = np.pad(img, [(0, r), (0, c), (0, 0)], 'constant')
gz, r2, c2 = DATA.patchify(img, scale=scale)
gz = net.get_model().predict(np.array(gz))
genz = DATA.reconstruct(gz, r2, c2, scale=1)
Image.fromarray(genz).save('bargen/' + f)
image_name = values.test_image
try:
img = cv2.imread(image_name)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
except cv2.error as e:
print("Bad image path check the name or path !!")
exit()
if not zoom:
r = DATA.patch_size - img.shape[0] % DATA.patch_size
c = DATA.patch_size - img.shape[1] % DATA.patch_size
img = np.pad(img, [(0, r), (0, c), (0, 0)], 'constant')
r = DATA.patch_size - img.shape[0] % DATA.patch_size
c = DATA.patch_size - img.shape[1] % DATA.patch_size
img = np.pad(img, [(0,r),(0,c),(0,0)] , 'constant')
Image.fromarray(img).save("div2k_padded_test.png")
lr_img = cv2.resize(img , (int(img.shape[1]/2),int(img.shape[0]/2)) ,cv2.INTER_CUBIC)
Image.fromarray(lr_img).save("div2k_padded_test_lr.png")
p , r , c = DATA.patchify(lr_img,scale=2)
#p = np.array(p) / 255
p = np.array(p)
first_image_Y = DATA.reconstruct(DATA.training_patches_Y , r , c , scale=1)
first_image_X = DATA.reconstruct(DATA.training_patches_2x , r , c , scale=2)
Image.fromarray(first_image_X).save("first_image_X.jpeg")
Image.fromarray(first_image_Y).save("first_image_Y.jpeg")
for i in range(chk+1,tryout):
print("tryout no: ",i)
samplev = np.random.random_integers(0 , DATA.training_patches_2x.shape[0]-1 , sample)
normalised_2x = DATA.training_patches_2x[samplev]
normalised_Y = DATA.training_patches_Y[samplev]
for i in range(chk + 1, tryout):
gen_acc = 0.0
dic_acc = 0.0
print("tryout no: ", i)
# samplev = np.random.random_integers(0 , x.shape[0]-1 , sample)
if adv_lambda > 0.0:
net.fit_discriminator(x, x2, x4, x8, batch_size, epoch=epochs)
net.fit(x, x2, x4, x8, batch_size, epoch=epochs)
net.get_model().save_weights('model_iter' + str(i) + '.h5')
g = net.predict(np.array(p))
gen2x = DATA.reconstruct(g[0], r, c, scale=4)
gen4x = DATA.reconstruct(g[1], r, c, scale=2)
gen8x = DATA.reconstruct(g[2], r, c, scale=1)
d = 'Results/' + str(i)
if not os.path.isdir(d):
os.mkdir(d)
Image.fromarray(gen2x).save(d + "/test_2x_gen_.png")
Image.fromarray(gen4x).save(d + "/test_4x_gen_.png")
Image.fromarray(gen8x).save(d + "/test_8x_gen_.png")
print("Reconstruction Gain:", PSNRLossnp(img, gen8x))
else:
if zoom:
gz, r2, c2 = DATA.patchify(img, scale=8)
gz = net.get_model().predict(np.array(gz))[2]
genz = DATA.reconstruct(gz, r2, c2, scale=1)
Image.fromarray(genz).save("test_image_zoomed_8x.png")