def get_pred_img_pixelwise(filename, image_idx, datatype, model, PIXEL_DEPTH,
NEW_DIM_TRAIN, prediction_test_dir):
i = image_idx
# Specify the path of the
if (datatype == 'train'):
imageid = "satImage_%.3d" % image_idx
image_filename = filename + imageid + ".png"
elif (datatype == 'test'):
imageid = "/test_%d" % i
image_filename = filename + imageid + imageid + ".png"
else:
print('Error: Enter test or train')
# loads the image in question
img = Image.open(image_filename)
output_prediction = get_prediction_pixel(img, model,
NEW_DIM_TRAIN) #(1,224,224)
output_prediction = np.transpose(output_prediction,
(1, 2, 0)) #(224,224,1)
predict_img = np.asarray(output_prediction)
# Changes into a 3D array, to easier turn into image
predict_img_3c = np.zeros((predict_img.shape[0], predict_img.shape[1], 3),
dtype=np.uint8)
predict_img8 = np.squeeze(img_float_to_uint8(predict_img, PIXEL_DEPTH))
predict_img8[predict_img8 >= 128] = 255
predict_img8[predict_img8 < 128] = 0
predict_img_3c[:, :, 0] = predict_img8
predict_img_3c[:, :, 1] = predict_img8
predict_img_3c[:, :, 2] = predict_img8
imgpred = Image.fromarray(predict_img_3c)
imgpredict = imgpred.resize((608, 608))
return imgpredict, img
def get_predictionimage(filename, image_idx, datatype, model, IMG_PATCH_SIZE,
PIXEL_DEPTH):
i = image_idx
# Specify the path of the
if (datatype == 'train'):
imageid = "satImage_%.3d" % image_idx
image_filename = filename + imageid + ".png"
elif (datatype == 'test'):
imageid = "/test_%d" % i
image_filename = filename + imageid + imageid + ".png"
else:
print('Error: Enter test or train')
# loads the image in question
img = mpimg.imread(image_filename)
#data = [img_patches[i][j] for i in range(len(img_patches)) for j in range(len(img_patches[i]))]
output_prediction = get_prediction(img, model, IMG_PATCH_SIZE)
predict_img = label_to_img(img.shape[0], img.shape[1], IMG_PATCH_SIZE,
IMG_PATCH_SIZE, output_prediction)
# Changes into a 3D array, to easier turn into image
predict_img_3c = np.zeros((img.shape[0], img.shape[1], 3), dtype=np.uint8)
predict_img8 = img_float_to_uint8(predict_img, PIXEL_DEPTH)
predict_img_3c[:, :, 0] = predict_img8
predict_img_3c[:, :, 1] = predict_img8
predict_img_3c[:, :, 2] = predict_img8
imgpred = Image.fromarray(predict_img_3c)
return imgpred
def get_predictionimage_context(filename, image_idx, datatype, model,
IMG_PATCH_SIZE, CONTEXT_SIZE, PIXEL_DEPTH):
if (datatype == 'train'):
imageid = "satImage_%.3d" % image_idx
image_filename = filename + imageid + ".png"
elif (datatype == 'test'):
imageid = "/test_%d" % image_idx
image_filename = filename + imageid + imageid + ".png"
else:
print('Error: Enter test or train')
img = mpimg.imread(image_filename)
output_prediction = get_prediction_context(img, model, IMG_PATCH_SIZE,
CONTEXT_SIZE)
predict_img = label_to_img(img.shape[0], img.shape[1], IMG_PATCH_SIZE,
IMG_PATCH_SIZE, output_prediction)
# Changes into a 3D array, to easier turn into image
predict_img_3c = np.zeros((img.shape[0], img.shape[1], 3), dtype=np.uint8)
predict_img8 = img_float_to_uint8(predict_img, PIXEL_DEPTH)
predict_img_3c[:, :, 0] = predict_img8
predict_img_3c[:, :, 1] = predict_img8
predict_img_3c[:, :, 2] = predict_img8
imgpred = Image.fromarray(predict_img_3c)
return imgpred
def make_img_overlay(img, predicted_img, PIXEL_DEPTH):
w = img.shape[0]
h = img.shape[1]
color_mask = np.zeros((w, h, 3),
dtype=np.uint8) #samme størrelse som bildet
color_mask[:, :,
0] = predicted_img * PIXEL_DEPTH #0 eller 3 Endrer bare R i rgb, altså gjør bildet
img8 = img_float_to_uint8(img, PIXEL_DEPTH)
background = Image.fromarray(img8, 'RGB').convert("RGBA")
overlay = Image.fromarray(color_mask, 'RGB').convert("RGBA")
new_img = Image.blend(background, overlay, 0.2)
return new_img
def make_img_overlay_pixel(img, predicted_img, PIXEL_DEPTH):
w, h = img.size
predicted_img = np.asarray(predicted_img)
color_mask = np.zeros((w, h, 3), dtype=np.uint8)
# creating a mask of the predicted image
color_mask[:, :, 0] = predicted_img[:, :, 0]
img8 = img_float_to_uint8(img, PIXEL_DEPTH)
background = Image.fromarray(img8, 'RGB').convert("RGBA")
overlay = Image.fromarray(color_mask, 'RGB').convert("RGBA")
new_img = Image.blend(background, overlay, 0.2)
return new_img
def get_prediction_with_overlay_pixelwise(filename, image_idx, datatype, model,
PIXEL_DEPTH, NEW_DIM_TRAIN,
IMG_PATCH_SIZE):
i = image_idx
if (datatype == 'train'):
imageid = "satImage_%.3d" % image_idx
image_filename = filename + imageid + ".png"
elif (datatype == 'test'):
imageid = "/test_%d" % i
image_filename = filename + imageid + imageid + ".png"
else:
print('Error: Enter test or train')
img = Image.open(image_filename)
# Returns a matrix with a prediction for each pixel
output_prediction = get_prediction_pixel(img, model,
NEW_DIM_TRAIN) #(1,400,400)
output_prediction = np.transpose(output_prediction,
(1, 2, 0)) #(400,400,1)
predict_img_3c = np.zeros(
(output_prediction.shape[0], output_prediction.shape[1], 3),
dtype=np.uint8)
predict_img8 = np.squeeze(
img_float_to_uint8(output_prediction, PIXEL_DEPTH))
predict_img8[predict_img8 >= 128] = 255
predict_img8[predict_img8 < 128] = 0
predict_img_3c[:, :, 0] = predict_img8
predict_img_3c[:, :, 1] = predict_img8
predict_img_3c[:, :, 2] = predict_img8
newPred = label_to_img_unet(predict_img_3c.shape[0],
predict_img_3c.shape[1], IMG_PATCH_SIZE,
IMG_PATCH_SIZE, output_prediction, datatype)
imgpred = Image.fromarray(newPred)
oimg = make_img_overlay_pixel(img, imgpred, PIXEL_DEPTH)
return oimg, imgpred