def main(patch_size, n_classes, model_path, output_filename, output_mapname, test_id, bands, class_weights):
# model = get_model(n_channels=4)
# test_id = 'test'
# test_id = '24'
img = normalize(tiff.imread('../data/mband/{}.tif'.format(test_id)).transpose([1, 2, 0])) # make channels last
n_channels = len(bands)
if bands[0] == -1:
n_channels = 8
print("\t num_channels", n_channels)
img = img[:, :, bands]
model = get_model(n_classes, patch_size, n_channels, class_weights)
model.load_weights(model_path)
"""
# Apply PCA to the image
pca_path = "./pca.pkl"
# open a file, where you stored the pickled data
file = open(pca_path, 'rb')
pca = pickle.load(file)
file.close()
reshaped_img_m = img.reshape((img.shape[0]*img.shape[1], img.shape[2]))
reshaped_img_pca = pca.transform(reshaped_img_m)
img_pca = reshaped_img_pca.reshape((img.shape[0], img.shape[1],
reshaped_img_pca.shape[1]))
img = img_pca
"""
mask = predict(img, model, patch_sz=patch_size, n_classes=n_classes).transpose([2,0,1]) # make channels first
map = picture_from_mask(mask, 0.5)
tiff.imsave(output_filename, (255*mask).astype('uint8'))
tiff.imsave(output_mapname, map)
for i in range(1, 9):
cl = z_order[i]
for ch in range(3):
pict[ch,:,:][mask[cl,:,:] > threshold] = colors[cl][ch]
return pict
if __name__ == '__main__':
model = get_model()
weights_path = "weights/unet_weights_30_50ep.hdf5"
model.load_weights(weights_path)
print(weights_path)
for i in range(1,15):
test_id = str(i)
print('Predicting for data/sat/{}.tif'.format(test_id))
img = normalize(tiff.imread('data/sat/{}.tif'.format(test_id)))
print(img.shape)
for i in range(15):
if i == 0: # reverse first dimension
mymat = predict(img[::-1,:,:], model, patch_sz=PATCH_SZ, n_classes=N_CLASSES).transpose([2,0,1])
#print(mymat[0][0][0], mymat[3][12][13])
# print("Case 1",img.shape, mymat.shape)
elif i == 1: # reverse second dimension
temp = predict(img[:,::-1,:], model, patch_sz=PATCH_SZ, n_classes=N_CLASSES).transpose([2,0,1])
#print(temp[0][0][0], temp[3][12][13])
print("Case 2", temp.shape, mymat.shape)
mymat = np.mean( np.array([ temp[:,::-1,:], mymat ]), axis=0 )
elif i == 2: # transpose(interchange) first and second dimensions
temp = predict(img.transpose([1,0,2]), model, patch_sz=PATCH_SZ, n_classes=N_CLASSES).transpose([2,0,1])
#print(temp[0][0][0], temp[3][12][13])
all_sliced_tifs = [
file for file in all_sliced_tifs if file[-4:] == ".tif"
][0:3]
total_files_count = len(all_sliced_tifs)
for current_file_count, test_file in enumerate(all_sliced_tifs[0:]):
test_filename = test_file
print("...running inference for {}".format(test_filename))
test_file = "../../satellite-image/Planet.com/Planet_Data_Sliced/tif/all-sliced-tiff-850px/" + test_file
class_id = 4
# img = take_4bands(normalize(tiff.imread('data/mband/{}.tif'.format(test_id)).transpose([1,2,0]))) # make channels last
# img = tiff.imread(test_file).transpose([1,2,0])
# print (img)
img = tiff.imread(test_file).transpose([1, 2, 0])
img, _ = get_4bands(
normalize(tiff.imread(test_file).transpose(
[1, 2, 0]))) # make channels last
band_list = [0, 1, 2, 3] # to permute all the bands
perm_list = list(itertools.permutations(band_list))
print("...the shape of given image is: ", img.shape)
res_shape = tuple(
[8] + list(img.shape[0:2]) +
[N_CLASSES]) # [8] here comes from the 8 augmented images
results = np.zeros(shape=res_shape)
for i in range(8):
print("...running inference for augmentation {}".format(i))
# TODO: The average taken in the following way isn't mathematically correct. Refer to the issue in original repo
if i == 0: # reverse first dimension
mymat = predict(img[::-1, :, :],
model,
patch_sz=PATCH_SZ,
n_classes=N_CLASSES).transpose([2, 0, 1])
z_order = {1: 3, 2: 4, 3: 0, 4: 1, 5: 2}
pict = 255 * np.ones(shape=(3, mask.shape[1], mask.shape[2]),
dtype=np.uint8)
for i in range(1, 6):
cl = z_order[i]
for ch in range(3):
pict[ch, :, :][mask[cl, :, :] > threshold] = colors[cl][ch]
return pict
if __name__ == '__main__':
model = get_model()
model.load_weights(weights_path)
test_id = 'test'
img = normalize(
tiff.imread('data/mband/{}.tif'.format(test_id)).transpose(
[1, 2, 0])) # make channels last
for i in range(7):
if i == 0: # reverse first dimension
mymat = predict(img[::-1, :, :],
model,
patch_sz=PATCH_SZ,
n_classes=N_CLASSES).transpose([2, 0, 1])
#print(mymat[0][0][0], mymat[3][12][13])
print("Case 1", img.shape, mymat.shape)
elif i == 1: # reverse second dimension
temp = predict(img[:, ::-1, :],
model,
patch_sz=PATCH_SZ,
n_classes=N_CLASSES).transpose([2, 0, 1])
x0, x1 = i * dimpatch, (i + 1) * dimpatch
y0, y1 = j * dimpatch, (j + 1) * dimpatch
print("I:", i)
print("J:", j)
print("SZ:", dimpatch)
print("X0,X1:", x0, x1)
print("Y0,Y1:", y0, y1)
prediction[x0:x1, y0:y1, :] = patches_predict[k, :, :, :]
return prediction[:h, :w, :]
if __name__ == '__main__':
model = get_model()
model.load_weights(weights_path)
img = normalize(tiff.imread('data/mband/test.tif'))
print("IMG H", img.shape[0])
print("IMG W", img.shape[1])
print("IMG Channel", img.shape[2])
#img = np.expand_dims(img,)
xdimens = img.shape[0]
ydimens = img.shape[1]
s = (xdimens, ydimens)
extrachannel1 = np.zeros(s)
extrachannel2 = np.zeros(s)
extrachannel3 = np.zeros(s)
extrachannel4 = np.zeros(s)
extrachannel5 = np.zeros(s)
output = np.zeros((xdimens, ydimens, 8))
output[:, :, 0] = img[:, :, 0]
output[:, :, 1] = img[:, :, 1]
'output/planet_classtest.tif', 'output/result.tif', 'output/map.tif',
'output/planet_result.tif', 'output/planet_map.tif'
]
for file in overwrite_check:
if os.path.exists(file):
print(
'ERROR: file {0} already exists. Please rename or delete the following files before creating predictions:'
.format(file))
print(*overwrite_check, sep='\n')
sys.exit()
if planet_test == False:
image_id = 'test'
weights_path = 'weights/150_epoch_unet_weights.hdf5'
img = normalize(
tiff.imread('data/mband/{}.tif'.format(image_id)).transpose(
[1, 2, 0])) # make channels last
elif planet_test == True:
planet_imagedir = 'data/planet_training/predict/'
image_id = '20180412_143154_1003_1B_AnalyticMS'
# rearranging order for planet image no longer necessary now that it matches training data
img = normalize(
tiff.imread(planet_imagedir + '{}.tif'.format(image_id)))
# add 4 channels of 0 to array to predict planet image
# img_pad = ((0,0), (0,0), (0,4))
# img_fixed2 = np.pad(img, pad_width=img_pad, mode='constant', constant_values=0)
#trim the planet image to the same dimensions as training data
# img_fixed2 = img_fixed2[:848, :837, :]
# tiff.imsave('output/planet_classtest.tif', img)
# img = img_fixed2
tiff.imsave('output/img01.tif', img)
def segment_result(path):
li = [os.path.join(path, i) for i in os.listdir(path)]
model = get_model()
model.load_weights(weights_path)
count = 0
for i in li:
img = normalize(tiff.imread(i).transpose([1, 2,
0])) # make channels last
img1 = i.split("/")[-1]
# test_id = img1.split(".")[0]
print(img1)
# img = normalize(tiff.imread('data/new_test/{}.tif'.format(test_id)).transpose([1,2,0])) # make channels last
for i in range(7):
if i == 0: # reverse first dimension
mymat = predict(img[::-1, :, :],
model,
patch_sz=PATCH_SZ,
n_classes=N_CLASSES).transpose([2, 0, 1])
print("================ mat ===================")
print(mymat[0][0][0], mymat[3][12][13])
print("Case 1", img.shape, mymat.shape)
elif i == 1: # reverse second dimension
temp = predict(img[:, ::-1, :],
model,
patch_sz=PATCH_SZ,
n_classes=N_CLASSES).transpose([2, 0, 1])
print("================ mat ===================")
print(temp[0][0][0], temp[3][12][13])
print("Case 2", temp.shape, mymat.shape)
mymat = np.mean(np.array([temp[:, ::-1, :], mymat]), axis=0)
elif i == 2: # transpose(interchange) first and second dimensions
temp = predict(img.transpose([1, 0, 2]),
model,
patch_sz=PATCH_SZ,
n_classes=N_CLASSES).transpose([2, 0, 1])
print("================ mat ===================")
print(temp[0][0][0], temp[3][12][13])
print("Case 3", temp.shape, mymat.shape)
mymat = np.mean(np.array([temp.transpose(0, 2, 1), mymat]),
axis=0)
elif i == 3:
temp = predict(np.rot90(img, 1),
model,
patch_sz=PATCH_SZ,
n_classes=N_CLASSES)
print("================ mat ===================")
print(
temp.transpose([2, 0, 1])[0][0][0],
temp.transpose([2, 0, 1])[3][12][13])
print("Case 4", temp.shape, mymat.shape)
mymat = np.mean(np.array(
[np.rot90(temp, -1).transpose([2, 0, 1]), mymat]),
axis=0)
elif i == 4:
temp = predict(np.rot90(img, 2),
model,
patch_sz=PATCH_SZ,
n_classes=N_CLASSES)
print("================ mat ===================")
print(
temp.transpose([2, 0, 1])[0][0][0],
temp.transpose([2, 0, 1])[3][12][13])
print("Case 5", temp.shape, mymat.shape)
mymat = np.mean(np.array(
[np.rot90(temp, -2).transpose([2, 0, 1]), mymat]),
axis=0)
elif i == 5:
temp = predict(np.rot90(img, 3),
model,
patch_sz=PATCH_SZ,
n_classes=N_CLASSES)
print("================ mat ===================")
print(
temp.transpose([2, 0, 1])[0][0][0],
temp.transpose([2, 0, 1])[3][12][13])
print("Case 6", temp.shape, mymat.shape)
mymat = np.mean(np.array(
[np.rot90(temp, -3).transpose(2, 0, 1), mymat]),
axis=0)
else:
temp = predict(img,
model,
patch_sz=PATCH_SZ,
n_classes=N_CLASSES).transpose([2, 0, 1])
print("================ mat ===================")
print(temp[0][0][0], temp[3][12][13])
print("Case 7", temp.shape, mymat.shape)
mymat = np.mean(np.array([temp, mymat]), axis=0)
print(mymat[0][0][0], mymat[3][12][13])
map = picture_from_mask(mymat, 0.5)
mask = predict(img, model, patch_sz=PATCH_SZ,
n_classes=N_CLASSES).transpose(
[2, 0, 1]) # make channels first
# print("====================Mask=======================")
# print(mask)
# print("=================map============================")
map = picture_from_mask(mask, 0.5)
# print(map)
# tiff.imsave('data/test1/mask.tif', (255*mask).astype('uint8'))
# tiff.imsave('data/test1/result.tif', (255*mymat).astype('uint8'))
tiff.imsave('data/test_set/test_map/map.{}.tif'.format(img1), map)
count += 1
if item[1].isdigit() else float('inf'), item)) #
#sorted_dir_list0 = sorted_dir_list0[918:]
#for z in sorted_dir_list0:
if __name__ == '__main__':
model = get_model()
model.load_weights(weights_path)
#model = load_model('/home/lpe/Desktop/junk/dish_ilastik_34/output/tha_gewd_u_net_2.h5',custom_objects={'weighted_binary_crossentropy': weighted_binary_crossentropy})
#test_id = "01"
#img = normalize(tiff.imread(z)).transpose([1,0,2]) # make channels last
for z in sorted_dir_list0:
# img = normalize(tiff.imread(z))#.transpose([1,0,2])
#img = image_resize(img,height=1173)#.transpose([1,0,2])
print(z)
img = normalize(image_resize(cv2.imread(z), height=1173))
for i in range(7):
if i == 0: # reverse first dimension
mymat = predict(img[::-1, :, :],
model,
patch_sz=PATCH_SZ,
n_classes=N_CLASSES).transpose([0, 2, 1])
#print(mymat[0][0][0], mymat[3][12][13])
print("Case 1", img.shape, mymat.shape)
elif i == 1: # reverse second dimension
temp = predict(img[:, ::-1, :],
model,
patch_sz=PATCH_SZ,
n_classes=N_CLASSES).transpose([1, 0, 2])
#print(temp[0][0][0], temp[3][12][13])
print("Case 2", temp.shape, mymat.shape)
overwrite_check = [
'output/planet_classtest.tif', 'output/result.tif', 'output/map.tif',
'output/planet_result.tif', 'output/planet_map.tif'
]
for file in overwrite_check:
if os.path.exists(file):
print(
'ERROR: file {0} already exists. Please rename or delete the following files before creating predictions:'
.format(file))
print(*overwrite_check, sep='\n')
sys.exit()
planet_imagedir = 'data/planet_training/predict/'
image_id = '20180412_143154_1003_1B_AnalyticMS'
img = normalize(tiff.imread(planet_imagedir + '{}.tif'.format(image_id)))
# add 4 channels of 0 to array to predict planet image
# img_pad = ((0,0), (0,0), (0,4))
# img_fixed2 = np.pad(img, pad_width=img_pad, mode='constant', constant_values=0)
#trim the planet image to the same dimensions as training data
# img_fixed2 = img_fixed2[:848, :837, :]
# tiff.imsave('output/planet_classtest.tif', img)
# img = img_fixed2
for i in range(7):
if i == 0: # reverse first dimension
mymat = predict(img[::-1, :, :],
model,
patch_sz=PATCH_SZ,
n_classes=N_CLASSES)
print("Case 1", img.shape, mymat.shape)
