def load_annotations(self, image_index):
""" Load annotations for an image_index.
"""
if self.set_name == 'valid':
image_index = self.random_index[image_index]
annotations = utils.load_annotations(
image_index, self.set_name) # (x1, y1, x2, y2, label)
permuted_annotations = np.zeros((annotations.shape[0], 5))
if annotations.shape[0] > 0:
permuted_annotations[:, 0] = annotations[:, 1]
permuted_annotations[:, 1] = annotations[:, 0]
permuted_annotations[:, 2] = annotations[:, 3]
permuted_annotations[:, 3] = annotations[:, 2]
permuted_annotations[:, 4] = annotations[:, 4]
return permuted_annotations
def transcript_scores_mapping(data_source, transcript_dirname,
annotation_dirname, output_dirname):
subtitle_dir = os.path.join(root_dir, data_source, transcript_dirname)
annotation_dir = os.path.join(root_dir, data_source, annotation_dirname)
output_dir = os.path.join(root_dir, data_source, output_dirname)
utils.mkdir(output_dir)
for fname in os.listdir(subtitle_dir):
id = fname[:-4]
print("Processing for ", id)
subtitle_file = os.path.join(subtitle_dir, fname)
anno_file = os.path.join(annotation_dir, "{}.csv".format(id))
output_file = os.path.join(output_dir, "{}.json".format(id))
subtitles = utils.load_subtitles(subtitle_file)
annotations = utils.load_annotations(anno_file)
utters = analyze_timing(subtitles, annotations)
with open(output_file, 'w') as writer:
json.dump([obj.__dict__ for obj in utters], writer, indent=3)
parser.error("SCALE_SIZE must be >= CROP_SIZE")
if prm.num_samples <= 0: parser.error("NUM_SAMPLES must be > 0")
prm_str = 'Arguments:\n' + '\n'.join(
['{} {}'.format(k.upper(), v) for k, v in vars(prm).items()])
print(prm_str + '\n')
# set paths for model wts
vnet_wts_fp = 'wts/pretrained/{}/visibnet.model.npz'.format(prm.input_attr)
cnet_wts_fp = 'wts/pretrained/{}/coarsenet.model.npz'.format(prm.input_attr)
rnet_wts_fp = 'wts/pretrained/{}/refinenet.model.npz'.format(prm.input_attr)
################################################################################
# Load annotations
anns = ut.load_annotations('data/anns/demo_5k/test.txt')
anns = anns[np.random.RandomState(seed=prm.seed).permutation(len(anns))]
# Load data
proj_depth = []
proj_sift = []
proj_rgb = []
src_img = []
gt_vis = []
for i in range(prm.num_samples):
# Load point cloud
pcl_xyz = ld.load_points_xyz('data/' + anns[i, 0])
pcl_rgb = ld.load_points_rgb('data/' + anns[i, 1])
pcl_sift = ld.load_points_sift('data/' + anns[i, 2])
default="datasets/VOCdevkit/VOC2010/JPEGImages",
help="Path to the PASCAL VOC 2010 JPEG images")
args = parser.parse_args()
# Stats on the dataset:
obj_cnt = 0
bodypart_cnt = 0
processed = 0
skipped = 0
mat_filenames = os.listdir(args.annotation_folder)
print("Starting...")
# Iterate through the .mat files contained in path:
for idx, annotation_filename in enumerate(mat_filenames):
annotations = utils.load_annotations(
os.path.join(args.annotation_folder, annotation_filename))
image_filename = annotation_filename[:annotation_filename.rfind(
".")] + ".jpg" # PASCAL VOC image have .jpg format
obj_cnt += len(annotations["objects"])
# Show original image with its mask:
img = PIL.Image.open(os.path.join(args.images_folder, image_filename))
img_width, img_height = img.size
total_mask = np.zeros([img_height, img_width])
labelExists = False
for obj in annotations["objects"]:
obj_name = obj['class']
if obj_name not in OBJECTS_FILTER:
continue
beam_size = 10
use_hidden = False
vocab_file = "/home/spb61/coco2014_vocab.json"
beam_captions_dir = "/datadrive/val_beam_{}_states/".format(beam_size)
print("Loading beams from: {}".format(beam_captions_dir))
save_file = "/home/spb61/image_captions/outputs/vote_captions_scores_{}_bigram_precision.pickle".format(
beam_size) # File to save generated votes
annotations_dir = "/home/spb61/annotations/"
annotations_file = "captions_val2014.json"
map_file = "/home/spb61/val_image_id_to_idx.csv"
vocab = load_vocab(dict_file=vocab_file)
image_id_to_index, index_to_image_id, annotations_dict = load_annotations(
annotations_dir=annotations_dir,
annotations_file=annotations_file,
map_file=map_file)
print("Found annotatoins for {} images".format(len(image_id_to_index)))
assert (len(image_id_to_index) == len(annotations_dict.keys()))
if not LOAD_CAPTIONS:
print("Saving voted caps to: {}".format(save_file))
vote_captions = {}
images = os.listdir(beam_captions_dir)
print("To process: {}".format(len(images)))
for i, image in enumerate(tqdm(images)):
image_id = int(image.split('.')[0])
try:
caption_object = load_caption(image_id,