# Crop image of labels that shape of labels match shape of cropped image.
labels_true_cropped = labels_true[0: img.shape[0] - img.shape[0] % 500,
0: img.shape[1] - img.shape[1] % 500]
# Get pixel-wise scores for labels 'building', 'highway' and 'no_label'.
scores = lib.predict_pixelwise_class_scores_for_image_advanced_1(img=img_cropped,
net=net,
net_type='predicting',
channel_means=channel_means,
label_nr=3)
# Predict label for each pixel in 'img_cropped'by choosing the label that got the highest score.
labels_predicted = np.argmax(scores, 2).astype(np.uint8)
# Calculate statistics for measuring the goodness of the prediction.
stats_temp = lib.calculate_goodness_of_classification(labels_true=labels_true_cropped,
labels_predicted=labels_predicted,
labels_dict=labels_dict)
# Cast dict objects 'stats_temp' to a pd.DataFrame object and append it
# to 'stats_all' that stores statistics for all images of this test round.
stats_all = stats_all.append(pd.DataFrame.from_dict(stats_temp, orient='index').T)
# Plot predicted class labels over aerial view image.
lib.plot_sat_image_and_labels(img=img_cropped,
lab=labels_predicted,
alpha=0.5,
plot_img=False,
store_img=True,
plot_with='Image',
dir_name=RESULTS_ROOT,
file_name='test_image_' + area_name + RUN_NAME + str(number_of_training_steps),
img_test_FCN_ready = img_test_FCN_ready.transpose(2, 0, 1).reshape(1, 3, 500, 500)
labels_test_FCN_ready = labels_test.reshape(1, 1, 500, 500)
# Load image into FCN.
solver.net.blobs['data'].data[...] = img_test_FCN_ready
# Send test image trough the FCN.
solver.net.forward()
# Get predicted class labels for each pixel of this round's test image.
labels_predicted = solver.net.blobs['score'].data.argmax(1)[0, :, :].astype(np.uint8)
# Get statistics for classification success of this test image.
# Note that we will get an object of type 'dict' back.
stats_new_dict = lib.calculate_goodness_of_classification(labels_true=labels_test.astype(np.uint8),
labels_predicted=labels_predicted,
labels_dict=labels_dict
)
# Cast this dict object to a pd.DataFrame object.
stats_new_DataFrame = pd.DataFrame.from_dict(stats_new_dict, orient='index').T
# Append this pd.DataFrame object to the pd.DataFrame object 'stats_current_training'
# which stores the statistics of all test images of this test round.
stats_current_training = stats_current_training.append(stats_new_DataFrame, ignore_index=True)
# Average stats in 'stats_current_training' column-wise over all the images which were
# used in this test round and append this average to 'stats_all'.
stats_all = stats_all.append(stats_current_training.T.mean(axis=1), ignore_index=True)
# Save stats to disc:
labels_true_cropped = labels_true_cropped.astype(np.int64)
# Get pixel-wise scores for labels 'building', 'highway' and 'no_label'.
scores = lib.predict_pixelwise_class_scores_for_image_advanced_1(img=img_cropped,
net=net,
net_type='predicting',
channel_means=channel_means)
# Predict label for each pixel in 'img_cropped'.
labels_predicted = np.argmax(scores, 2)
# Calculate statistics for measuring the goodness of the prediction.
stats_temp = lib.calculate_goodness_of_classification(labels_true = labels_true_cropped,
labels_predicted = labels_predicted,
return_precision=True,
return_recall=True,
return_f1=True,
return_pixel_accuracy=False,
return_mean_accuracy=False,
return_mean_iu=False,
return_frequency_weighted_iu=False)
# Cast dict objects 'stats_temp' to a pd.DataFrame object and append it
# to 'stats_all' that stores statistics for all images of this test round.
stats_temp_Data_Frame = pd.DataFrame(stats_temp.values()).T
stats_temp_Data_Frame.columns = stats_temp.keys()
stats_all = stats_all.append(stats_temp_Data_Frame)
# Plot predicted class labels over aerial view image.
lib.plot_sat_image_and_labels(img=img_cropped,
lab_1=labels_predicted,
lab_2=None,
# Load image into FCN.
solver.net.blobs['data'].data[...] = img_test_FCN_ready
# Send test image trough the FCN.
solver.net.forward()
# Get predicted class labels for each pixel of this round's test image.
labels_predicted = solver.net.blobs['score'].data.argmax(1)[0, :, :]
# Get statistics for classification success of this test image.
# Note that we will get an object of type 'dict' back.
stats_new_dict = lib.calculate_goodness_of_classification(labels_true=labels_test.astype(np.int64),
labels_predicted=labels_predicted,
return_precision=True,
return_recall=True,
return_f1=True,
return_pixel_accuracy=False,
return_mean_accuracy=False,
return_mean_iu=False,
return_frequency_weighted_iu=False
)
# Cast this dict object to a pd.DataFrame object.
stats_new_DataFrame = pd.DataFrame(stats_new_dict.values()).T
stats_new_DataFrame.columns = stats_new_dict.keys()
# Append this pd.DataFrame object to the pd.DataFrame object 'stats_current_training'
# which stores the statistics of all test images of this test round.
stats_current_training = stats_current_training.append(stats_new_DataFrame, ignore_index=True)
# Average stats in 'stats_current_training' column-wise over all the images which were
# used in this test round and append this average to 'stats_all'.
labels_google_osm_bld[labels_google_osm[:, :, 0] != 255] = False
labels_google_osm_bld[labels_google_osm[:, :, 1] != 0] = False
labels_google_osm_bld[labels_google_osm[:, :, 2] != 0] = False
labels_google_osm_hgh = np.ones(shape=labels_google_osm.shape[0:2], dtype=bool)
labels_google_osm_hgh[labels_google_osm[:, :, 0] != 0] = False
labels_google_osm_hgh[labels_google_osm[:, :, 1] != 0] = False
labels_google_osm_hgh[labels_google_osm[:, :, 2] != 255] = False
labels_google_osm_out = np.zeros(shape=labels_google_osm.shape[0:2]).astype(np.uint8)
labels_google_osm_out[labels_google_osm_bld] = 1
labels_google_osm_out[labels_google_osm_hgh] = 2
# Calculate some basic statistics.
stats_temp = lib.calculate_goodness_of_classification(labels_true=labels_google_osm_out,
labels_predicted=labels_ISPRS,
labels_dict=labels_dict)
stats_all = stats_all.append(pd.DataFrame.from_dict(stats_temp, orient='index').T)
# lib.plot_true_and_pred_labels(true_lab=labels_ISPRS,
# prd_lab=labels_google_osm_out,
# plot_img=False,
# store_img=True,
# dir_name=STORE_DATA_ROOT + area_name + '/',
# file_name=area_name + "_osm_isprs_labels_overlay"
# )
# Calculate average statistics
stats_average = stats_all.T.mean(axis=1)
