def plot_non_mask_rate_iou(self, model, dataset, metric, epoch):
run_id = model.run_id
test_y = self.data_collection.test_y_dec
test_predict = self.data_collection.test_predict_dec
size = 512
test_y = test_y[:size]
test_predict = test_predict[:size]
xs = masks_rate(test_y)
xs = xs.reshape([-1])
xs /= 255
ys = np.array([
Metrics.miou(true, predict)
for true, predict in zip(test_y, test_predict)
])
dots = np.array([[x, y] for x, y in zip(xs, ys)])
self.plot.scatter_2d(dots,
title=f'test set mask rate and iou',
path=path_join(
PLOT_PATH, run_id,
f'test_set_mask_rate_iou/({epoch}).png'),
x_label='mask_rate',
y_label='iou')
def test_TGS_salt_metric():
print(f'collect train mask images')
train_mask_images, _, _ = collect_images(TRAIN_MASK_PATH)
train_mask_images = train_mask_images.reshape([-1, 101, 101])
idx = 10
size = 10
images = train_mask_images[idx:idx + size]
metric_images = []
for a in images:
for b in images:
a = a.reshape([101, 101, 1])
b = b.reshape([101, 101, 1])
zero_channel = np.zeros([101, 101, 1])
rgb_image = np.concatenate([a, b, zero_channel], axis=2)
metric_images += [rgb_image]
metric_images = np.array(metric_images)
plot.plot_image_tile(metric_images, title='test_metric', column=size)
metric_score = []
for a in images:
for b in images:
metric_score += [Metrics.TGS_salt_score(a, b)]
metric_score = np.array(metric_score).reshape([size, size])
print(metric_score)
def fold_score(self, k=5):
models = []
train_TGS_scores = []
valid_TGS_scores = []
train_mious = []
valid_mious = []
for i in range(k):
path = f'./instance/TGS_salt/SS/baseline/fold_{i}'
model = self.load_model(path)
models += model
datas = self.prepare_set(k=5, index=i)
train_x_enc, train_y_enc, valid_x_enc, valid_y_enc = self.encode_datas(
datas)
train_predict = model.predict(train_x_enc)
valid_predict = model.predict(valid_y_enc)
train_TGS_score = Metrics.TGS_salt_score(train_y_enc,
train_predict)
valid_TGS_score = Metrics.TGS_salt_score(valid_y_enc,
valid_predict)
train_iou_score = Metrics.miou(train_y_enc, train_predict)
valid_iou_score = Metrics.miou(valid_y_enc, valid_predict)
train_TGS_scores += [train_TGS_score]
valid_TGS_scores += [valid_TGS_score]
train_mious += [train_iou_score]
valid_mious += [valid_iou_score]
def print_score(scores):
for i, score in enumerate(scores):
print(i, score)
print(f'train TGS score')
print_score(train_TGS_scores)
print(f'valid TGS score')
print_score(valid_TGS_scores)
print(f'train ious')
print_score(train_mious)
print(f'valid ious')
print_score(valid_mious)
def log_score(self, k, index):
baseline = self.model
datas = self.prepare_set(k, index)
train_x_enc, train_y_enc, valid_x_enc, valid_y_enc = self.encode_datas(
datas)
train_predict = baseline.predict(train_x_enc)
valid_predict = baseline.predict(valid_x_enc)
train_TGS_score = Metrics.TGS_salt_score(train_y_enc, train_predict)
valid_TGS_score = Metrics.TGS_salt_score(valid_y_enc, valid_predict)
train_iou_score = Metrics.miou(train_y_enc, train_predict)
valid_iou_score = Metrics.miou(valid_y_enc, valid_predict)
print(f'train TGS score = {train_TGS_score}\n'
f'test TGS score = {valid_TGS_score}\n'
f'train miou = {train_iou_score}\n'
f'test miou = {valid_iou_score}\n'
# f'train loss = {train_loss}\n'
# f'test loss = {test_loss}\n'
)
def test_set_sorted_by_mask_rate(self):
pipe = self.load_baseline()
baseline = pipe.model
train_predict = baseline.predict(pipe.train_x)
train_predict = mask_label_encoder.from_label(train_predict)
test_predict = baseline.predict(pipe.valid_x)
test_predict = mask_label_encoder.from_label(test_predict)
train_loss = baseline.metric(pipe.train_x, pipe.train_y_encode)
test_loss = baseline.metric(pipe.valid_x, pipe.valid_y_encode)
train_proba = baseline.predict_proba(pipe.train_x)
test_proba = baseline.predict_proba(pipe.valid_x)
train_score = Metrics.TGS_salt_score(pipe.train_y, train_predict)
test_score = Metrics.TGS_salt_score(pipe.valid_y, test_predict)
train_mask_rate = masks_rate(pipe.train_y)
test_mask_rate = masks_rate(pipe.valid_y)
train_miou = Metrics.miou(pipe.train_y, train_predict)
test_miou = Metrics.miou(pipe.valid_y, test_predict)
train_ious = Metrics.iou(pipe.train_y, train_predict)
test_ious = Metrics.iou(pipe.valid_y, test_predict)
test_x = pipe.valid_x
test_y = pipe.valid_y
test_predict = test_predict
test_mask_rate = test_mask_rate
zipped = zip(test_x, test_y, test_predict, test_mask_rate)
sort = list(sorted(zipped, key=lambda a: a[3]))
test_x, test_y, test_predict, test_mask_rate = zip(*sort)
test_x = np.array(test_x)
test_y = np.array(test_y)
test_predict = np.array(test_predict)
test_mask_rate = np.array(test_mask_rate)
test_x_image = test_x[:, :, :, 0]
test_x_image = test_x_image.reshape([-1, 101, 101, 1])
test_y = test_y.reshape([-1, 101, 101, 1])
test_predict = test_predict.reshape([-1, 101, 101, 1])
# plot_path = f'./matplot/baseline'
tile_image = self.scramble_column(test_x_image, test_y, test_predict)
plot.plot_image_tile(
tile_image,
title='all',
path=f'./matplot/baseline/test_set_sorted_by_mask_rate.png')
def update_data(self, model, dataset, metric, epoch):
def decode(x):
return x * 255
self.train_loss = metric
# train_x_enc, train_y_enc = dataset.next_batch(500)
# self.train_x_enc = train_x_enc
# self.train_x_dec = decode(self.train_x_enc)
# self.train_y_enc = train_y_enc
# self.train_y_dec = decode(self.train_y_enc)
# self.train_predict_enc = model.predict(self.train_x_enc)
# self.train_predict_dec = decode(self.train_predict_enc)
self.test_predict_enc = model.predict(self.test_x_enc)
self.test_predict_dec = decode(self.test_predict_enc)
# self.train_score = Metrics.TGS_salt_score(self.train_y_enc, self.train_predict_enc)
self.test_score = Metrics.TGS_salt_score(self.test_y_enc,
self.test_predict_enc)
def log_split_mask_rate_score(self):
pipe = self.load_baseline()
baseline = pipe.model
helper = pipe.data_helper
train_set = helper.train_set_non_empty_mask_with_depth_image_under_1p
train_set.x_keys = ['x_with_depth']
x, y = train_set.full_batch()
predict = baseline.predict(x)
score_under_1 = Metrics.TGS_salt_score(
y, mask_label_encoder.from_label(predict))
iou_score_under_1 = Metrics.miou(
y, mask_label_encoder.from_label(predict))
train_set = helper.train_set_non_empty_mask_with_depth_image_under_5p
train_set.x_keys = ['x_with_depth']
x, y = train_set.full_batch()
predict = baseline.predict(x)
score_under_5 = Metrics.TGS_salt_score(
y, mask_label_encoder.from_label(predict))
iou_score_under_5 = Metrics.miou(
y, mask_label_encoder.from_label(predict))
train_set = helper.train_set_non_empty_mask_with_depth_image_under_10p
train_set.x_keys = ['x_with_depth']
x, y = train_set.full_batch()
predict = baseline.predict(x)
score_under_10 = Metrics.TGS_salt_score(
y, mask_label_encoder.from_label(predict))
iou_score_under_10 = Metrics.miou(
y, mask_label_encoder.from_label(predict))
train_set = helper.train_set_non_empty_mask_with_depth_image_under_20p
train_set.x_keys = ['x_with_depth']
x, y = train_set.full_batch()
predict = baseline.predict(x)
score_under_20 = Metrics.TGS_salt_score(
y, mask_label_encoder.from_label(predict))
iou_score_under_20 = Metrics.miou(
y, mask_label_encoder.from_label(predict))
train_set = helper.train_set_non_empty_mask_with_depth_image_upper_1p
train_set.x_keys = ['x_with_depth']
x, y = train_set.full_batch()
predict = baseline.predict(x)
score_upper_1 = Metrics.TGS_salt_score(
y, mask_label_encoder.from_label(predict))
iou_score_upper_1 = Metrics.miou(
y, mask_label_encoder.from_label(predict))
train_set = helper.train_set_non_empty_mask_with_depth_image_upper_5p
train_set.x_keys = ['x_with_depth']
x, y = train_set.full_batch()
predict = baseline.predict(x)
score_upper_5 = Metrics.TGS_salt_score(
y, mask_label_encoder.from_label(predict))
iou_score_upper_5 = Metrics.miou(
y, mask_label_encoder.from_label(predict))
train_set = helper.train_set_non_empty_mask_with_depth_image_upper_10p
train_set.x_keys = ['x_with_depth']
x, y = train_set.full_batch()
predict = baseline.predict(x)
score_upper_10 = Metrics.TGS_salt_score(
y, mask_label_encoder.from_label(predict))
iou_score_upper_10 = Metrics.miou(
y, mask_label_encoder.from_label(predict))
train_set = helper.train_set_non_empty_mask_with_depth_image_upper_20p
train_set.x_keys = ['x_with_depth']
x, y = train_set.full_batch()
predict = baseline.predict(x)
score_upper_20 = Metrics.TGS_salt_score(
y, mask_label_encoder.from_label(predict))
iou_score_upper_20 = Metrics.miou(
y, mask_label_encoder.from_label(predict))
print(f'\n'
f' TGS score\n'
f'under 1 {score_under_1}\n'
f'under 5 {score_under_5}\n'
f'under 10 {score_under_10}\n'
f'under 20 {score_under_20}\n'
f'upper 1 {score_upper_1}\n'
f'upper 5 {score_upper_5}\n'
f'upper 10 {score_upper_10}\n'
f'upper 20 {score_upper_20}\n'
f'\n'
f'\n'
f' iou score\n'
f'under 1 {iou_score_under_1}\n'
f'under 5 {iou_score_under_5}\n'
f'under 10 {iou_score_under_10}\n'
f'under 20 {iou_score_under_20}\n'
f'upper 1 {iou_score_upper_1}\n'
f'upper 5 {iou_score_upper_5}\n'
f'upper 10 {iou_score_upper_10}\n'
f'upper 20 {iou_score_upper_20}\n'
f'\n')