def generate_images(image_path="train/1.png",net_path='net.h5',full=False):
image = scipy.ndimage.imread(image_path)
weights_h5 = h5py.File(net_path, 'r')
weights = []
maxouts = []
for i in range(99):
try:
weights.append(np.array(weights_h5[str(i)],dtype=np.float32))
try:
maxouts.append((weights_h5[str(i)].attrs['maxpool_x'],weights_h5[str(i)].attrs['maxpool_y']))
except:
maxouts.append(-1)
except:
print("Ending on:",i)
break
output = h5py.File('output.h5', 'w')
if not full:
for j in range(len(weights)):
layer = output.create_group(str(j))
for i in range(weights[j].shape[1]):
im = deconv_image(image,weights,j+1,i,maxouts)
output[str(j)].create_dataset(str(i), data=im)
else:
for i in range(len(weights)):
im = deconv_image(image,weights,i+1,'all',maxouts)
plt.imshow(im)
output.create_dataset(str(i), data=im)
plt.show()
def display_images(image_path="train/1.png",net_path='net.h5'):
image = scipy.ndimage.imread(image_path)
weights_h5 = h5py.File(net_path, 'r')
weights = []
maxouts = []
for i in range(99):
try:
weights.append(np.array(weights_h5[str(i)],dtype=np.float32))
try:
maxouts.append((weights_h5[str(i)].attrs['maxpool_x'],weights_h5[str(i)].attrs['maxpool_y']))
except:
maxouts.append(-1)
except:
print("Ending on:",i)
break
width = np.min([weights[j].shape[0] for j in range(len(weights))])
plt.figure(figsize = (len(weights),width))
gs1 = gridspec.GridSpec( len(weights),width)
gs1.update(wspace=0.001, hspace=0.001)
images = []
idx = 0
for j in range(len(weights)):
for i in range(width):
im = deconv_image(image,weights,j+1,i,maxouts)
ax1 = plt.subplot(gs1[idx])
plt.axis('on')
ax1.set_xticklabels([])
ax1.set_yticklabels([])
ax1.set_aspect('equal')
plt.imshow(im)
idx+=1
plt.show()
