def __getitem__(self, index):
start = self.batch_size * index
end = min(start + self.batch_size, len(self.match) + len(self.unmatch))
size = end - start
assert size > 0
def _read_for_training_expand_dim(p):
return np.expand_dims(read_for_training(p), 0).astype(K.floatx())
i_start, i_end = start // 2, int(start // 2 + np.ceil(size / 2))
unroll_match = [[m[0], m[1], u[0], u[1]] for m, u in zip(
self.match[i_start:i_end], self.unmatch[i_start:i_end])]
unroll_match = [i for item in unroll_match for i in item]
unroll_train_images = np.concatenate(
self.p.map(_read_for_training_expand_dim, unroll_match), 0)
#print("\nunroll_train_images shape: {}\n".format(unroll_train_images.shape))
unroll_train_images = preprocess_image(unroll_train_images)
## a[0], b[0], a[1], b[1], a[2], b[2], a[3], a[4] ...
a_indices = list(range(0, len(unroll_train_images), 2))
a = unroll_train_images[a_indices]
b_indices = list(range(1, len(unroll_train_images), 2))
b = unroll_train_images[b_indices]
c = np.zeros((size, 1), dtype=K.floatx())
for i in range(0, size, 2):
c[i, 0] = 1 # This is a match
c[i + 1, 0] = 0 # Different whales
return [a, b], c
def process_train_images_with_branch_model(branch_model, fold):
cache_path = CACHE_PATH + 'branch_res_train_fold_{}.pklz'.format(fold)
if not os.path.isfile(cache_path):
print('Go branch model train...')
start_time = time.time()
bboxes = get_boxes()
CLASSES = get_classes_array()
tagged = get_tagged_data()
train_ids = []
labels = dict()
train_size = len(tagged)
missed = 0
# train_size = 10
train_res = []
i = 0
for image_id in sorted(list(tagged.keys())):
clss = tagged[image_id]
# print('Go for: {}'.format(image_id))
f = expand_path(image_id)
lbl = CLASSES.index(clss)
if image_id not in bboxes:
print('Missed bbox for {}!'.format(image_id))
missed += 1
continue
bb = bboxes[image_id]
labels[image_id] = lbl
train_ids.append(image_id)
img = read_cropped_image(f,
bb[0],
bb[1],
bb[2],
bb[3],
False,
img_shape=(BOX_SIZE, BOX_SIZE, 3))
img = np.expand_dims(img, axis=0)
img = preprocess_image(img.astype(np.float32))
preds = branch_model.predict(img)
train_res.append(preds[0])
i += 1
train_res = np.array(train_res)
train_ids = np.array(train_ids)
print('Train preds: {}'.format(train_res.shape))
print('Missed bboxes: {}'.format(missed))
print('Read train time: {:.0f} sec'.format(time.time() - start_time))
save_in_file((train_res, train_ids, labels), cache_path)
else:
print('Restore train from cache: {}'.format(cache_path))
train_res, train_ids, labels = load_from_file(cache_path)
return train_res, train_ids, labels
def __getitem__(self, index):
start = self.batch_size * index
size = min(len(self.data) - start, self.batch_size)
def _read_for_validation_expand_dim(p):
return np.expand_dims(read_for_validation(p), 0).astype(K.floatx())
res = self.p.map(_read_for_validation_expand_dim,
self.data[start:start + size])
res = np.concatenate(res, 0)
res = preprocess_image(res)
return res
def process_tst_images_with_branch_model(branch_model, fold):
cache_path = CACHE_PATH + 'branch_res_test_fold_{}.pklz'.format(fold)
if not os.path.isfile(cache_path):
print('Go branch model test...')
start_time = time.time()
bboxes = get_boxes()
missed = 0
test_df = pd.read_csv(INPUT_PATH + 'sample_submission.csv')
test_ids = []
test_size = len(test_df)
# test_size = 10
test_res = []
i = 0
for index, row in test_df.iterrows():
# print('Go for: {}'.format(row['Image']))
f = INPUT_PATH + 'test/' + row['Image']
image_id = row['Image']
if image_id in bboxes:
bb = bboxes[image_id]
else:
print('Missed bbox for {}!'.format(image_id))
missed += 1
img = read_single_image(f)
bb = (0, 0, img.shape[1], img.shape[0])
img = read_cropped_image(f,
bb[0],
bb[1],
bb[2],
bb[3],
False,
img_shape=(BOX_SIZE, BOX_SIZE, 3))
img = np.expand_dims(img, axis=0)
img = preprocess_image(img.astype(np.float32))
preds = branch_model.predict(img)
test_res.append(preds[0])
test_ids.append(row['Image'])
i += 1
test_res = np.array(test_res)
test_ids = np.array(test_ids)
print('Test preds: {}'.format(test_res.shape))
print('Missed bboxes: {}'.format(missed))
print('Read test time: {:.0f} sec'.format(time.time() - start_time))
save_in_file((test_res, test_ids), cache_path)
else:
print('Restore test from cache: {}'.format(cache_path))
test_res, test_ids = load_from_file(cache_path)
return test_res, test_ids