def post(self, picture):
# load image from request
imgdata = base64.b64decode(picture.replace('&', '/'))
with open('drawings/drawing.jpg', 'wb') as f:
f.write(imgdata)
X = load_test_image()
# load optimized theta arrays
theta1_opt = np.load('theta1_opt_76-95.npy')
theta2_opt = np.load('theta2_opt_76-95.npy')
# make prediction and get percentages
pred, percentages = analize_single(theta1_opt, theta2_opt, X)
class_ids, ids_to_class = get_class_ids()
prediction = ids_to_class[str(pred[0])]
slice = dict()
slice.update({'prediction': prediction})
i = 0
for percentage in percentages:
slice.update({ids_to_class[str(i)]: percentage})
i += 1
return slice, 201
def test_histogram_add_fixed_color():
"""
Test Histogram.add_fixed_color
"""
width_A, height_A, input_pixels_A = utils.load_test_image('flower')
width_B, height_B, input_pixels_B = utils.load_test_image('flower-huechange-1')
assert width_A == width_B
width, height = width_A, height_A
attr = liq.Attr()
hist = liq.Histogram(attr)
image_A = attr.create_rgba(input_pixels_A, width, height, 0)
hist.add_image(attr, image_A)
image_B = attr.create_rgba(input_pixels_B, width, height, 0)
hist.add_image(attr, image_B)
FIXED_COLORS = [
liq.Color(255, 0, 0, 255), # red
liq.Color(0, 0, 255, 255), # blue
liq.Color(128, 0, 128, 255), # purple
liq.Color(255, 255, 0, 255), # yellow
]
for fixedColor in FIXED_COLORS:
hist.add_fixed_color(fixedColor, 0)
result = hist.quantize(attr)
result_palette = result.get_palette()
# Check that we have a decently-sized palette
assert len(result_palette) > 128
# Try remapping both images -- make sure we don't get an exception
# or something
result.remap_image(image_A)
result.remap_image(image_B)
# Check that the fixed colors are present in the output palette
for fixedColor in FIXED_COLORS:
assert fixedColor in result_palette
def image_callback(value, image):
# Load the map as a bytearray
width, height, input_pixels = utils.load_test_image(value)
buffer = bytearray(width * height)
for i in range(0, len(input_pixels), 4):
buffer[i // 4] = input_pixels[i]
# Test both the getter and setter methods
image.importance_map = buffer
with pytest.raises(AttributeError):
image.importance_map
def image_callback(value, image):
# Load the background as an Image
width, height, input_pixels = utils.load_test_image(value)
attr = liq.Attr()
background = attr.create_rgba(input_pixels, width, height, 0)
backgrounds.append(background)
# Test both the getter and setter methods
image.background = background
with pytest.raises(AttributeError):
image.background
def test_histogram_basic():
"""
Basic test of Histogram:
- __init__()
- add_image()
- quantize()
"""
# Testing with three input images
width_A, height_A, input_pixels_A = utils.load_test_image('flower')
width_B, height_B, input_pixels_B = utils.load_test_image('flower-huechange-1')
width_C, height_C, input_pixels_C = utils.load_test_image('flower-huechange-2')
assert width_A == width_B == width_C
assert height_A == height_B == height_C
width, height = width_A, height_A
attr = liq.Attr()
hist = liq.Histogram(attr)
image_A = attr.create_rgba(input_pixels_A, width, height, 0)
hist.add_image(attr, image_A)
image_B = attr.create_rgba(input_pixels_B, width, height, 0)
hist.add_image(attr, image_B)
image_C = attr.create_rgba(input_pixels_C, width, height, 0)
hist.add_image(attr, image_C)
result = hist.quantize(attr)
# Check that we have a decently-sized palette
assert len(result.get_palette()) > 128
# Try remapping all three images -- make sure we don't get an
# exception or something
result.remap_image(image_A)
result.remap_image(image_B)
result.remap_image(image_C)
def test_attr_copy():
"""
Test Attr.copy()
"""
width, height, input_pixels = utils.load_test_image('flower')
attr = liq.Attr()
attr.max_colors = 88
attr.min_posterization = 3
attr.min_quality = 55
attr2 = attr.copy()
assert attr2.max_colors == 88
assert attr2.min_posterization == 3
assert attr2.min_quality == 55
def test_buffer_too_small_error():
"""
Trigger LIQ_BUFFER_TOO_SMALL and ensure that liq.BufferTooSmallError
is raised
"""
# Load the background as an Image
width, height, input_pixels = utils.load_test_image('test-card')
attr = liq.Attr()
background = attr.create_rgba(input_pixels, width, height, 0)
def image_callback(value, image):
# The image is too large, so using it as a background should fail
with pytest.raises(liq.BufferTooSmallError):
image.background = background
utils.try_multiple_values('alpha-gradient', [None],
image_callback=image_callback)
parser.add_argument("target_ds")
args = parser.parse_args()
print("model loaded.")
d_img = path.join(args.target_ds, "images") + "/*"
d_mask = path.join(args.target_ds, "masks") + "/*"
img_paths = sorted(glob.glob(d_img))
mask_paths = sorted(glob.glob(d_mask))
j, wrong_count = 0, 0
y_pred, y_true = [], []
print(img_paths)
for img_path, mask_path in zip(img_paths, mask_paths):
img = load_test_image(img_path, dsize=(IMG_HEIGHT, IMG_WIDTH))
img_hsv = rgb2hsv(img, normalization=True)
mask = pb_predict_mask(sess, img) > CLOUD_THRES
#mask = pb_predict_window_mask(sess, img_hsv, window_size = SLIDING_WINDOW_SIZE) > CLOUD_THRES
#mask = np.zeros_like(img_hsv[...,0])
mask_pct = np.sum(mask) / (mask.shape[0] * mask.shape[1]) * 100
gt_mask = load_mask(mask_path,
dsize=(IMG_HEIGHT, IMG_WIDTH)) / 255.0 > CLOUD_THRES
gt_mask_pct = np.sum(gt_mask) / (gt_mask.shape[0] * gt_mask.shape[1]) * 100
y_pred.append(mask_pct)
y_true.append(gt_mask_pct)
if pct_to_label(mask_pct) != pct_to_label(gt_mask_pct):
wrong_count += 1
print("Wrong Prediction. {} has around {}% of sky area. Predicted:{}%".
format(img_path, PCT_LVL[pct_to_label(gt_mask_pct)],
def test_histogram_add_colors():
"""
Test Histogram.add_colors(), as well as the HistogramEntry class
"""
# First, quantize flower-huechange-1.jpg on its own
width, height, input_pixels = utils.load_test_image('flower-huechange-1')
attr = liq.Attr()
other_image = attr.create_rgba(input_pixels, width, height, 0)
result = other_image.quantize(attr)
result_pixels = result.remap_image(other_image)
result_palette = result.get_palette()
# ~
# Create a list of HistogramEntrys for it
entries = []
for i, color in enumerate(result_palette):
count = result_pixels.count(i)
entry = liq.HistogramEntry(color, count)
entries.append(entry)
# Test HistogramEntry getters
assert entry.color == color
assert entry.count == count
# Test setters
entry.color = result_palette[0]
entry.count = 50
assert entry.color == result_palette[0]
assert entry.count == 50
# (Set properties back to what they should be)
entry.color = color
entry.count = count
assert entries
# ~
# Set up a Histogram
attr = liq.Attr()
hist = liq.Histogram(attr)
# Add flower.jpg as an image
width, height, input_pixels = utils.load_test_image('flower')
image = attr.create_rgba(input_pixels, width, height, 0)
hist.add_image(attr, image)
# Add the HistogramEntrys for flower-huechange-1.jpg
hist.add_colors(attr, entries, 0)
# And quantize
result = hist.quantize(attr)
# ~
# Check that we have a decently-sized palette
assert len(result.get_palette()) > 128
# Try remapping both images -- make sure we don't get an exception
# or something
result.remap_image(image)
result.remap_image(other_image)
# for i in range(len(x_train)):
# pre = np.squeeze(model.predict(x_train[i:i+1]), axis=(0,3))
# train_patch_pre_label = np.array(list(get_patch_label_from_array(pre,th=0.21)))
# ground_truth_label = np.array(list(get_patch_label_from_array(y_train[i,:,:,0],th=0.21)))
# train_acc.append(np.mean(np.equal(train_patch_pre_label[:,2],ground_truth_label[:,2])))
#
# # valid acc
# for i in range(len(x_valid)):
# pre = np.squeeze(model.predict(x_valid[i:i+1]), axis=(0,3))
# train_patch_pre_label = np.array(list(get_patch_label_from_array(pre,th=0.21)))
# ground_truth_label = np.array(list(get_patch_label_from_array(y_valid[i,:,:,0],th=0.21)))
# train_acc.append(np.mean(np.equal(train_patch_pre_label[:,2],ground_truth_label[:,2])))
# make submission
from utils import load_test_image
import cv2 as cv
test_mat,test_img_id = load_test_image()
th = 0.21
f = open('submission.csv','w')
f.write('id,prediction\n')
for i in range(len(test_img_id)):
print(i)
img = np.expand_dims(cv.resize(test_mat[i], dsize=(400, 400), interpolation=cv.INTER_CUBIC),axis=0)
pre = np.squeeze(model.predict(img),axis=(0,3))
pre_resize = cv.resize(pre,dsize=(test_mat[0].shape[0],test_mat[0].shape[1]),interpolation=cv.INTER_CUBIC)
train_patch_pre_label = list(get_patch_label_from_array(pre_resize,th=th))
for line in train_patch_pre_label:
final_string = "{:03d}_{}_{},{}".format(test_img_id[i], line[0], line[1], line[2])
f.write(final_string+'\n')