ref_im = scene_list[j].strip()
query_im = scene_list[0].strip()
scene_dir = ref_im.split('/')
scene_dir = '/'.join(scene_dir[:-1])
# completely define various path
query_im = os.path.join(args.root, query_im)
query_im_ = query_im
ref_im = os.path.join(args.root, ref_im)
scene_dir = os.path.join(args.root, scene_dir)
H_file = os.path.join(args.root, scene_dir, 'H_1_%s' % str(j + 1))
query_im = imreadth(query_im)
hA, wA = query_im.shape[-2:]
query_im = resize(normalize(query_im), args.image_size,
scale_factor)
hA_, wA_ = query_im.shape[-2:]
ref_im = imreadth(ref_im)
hB, wB = ref_im.shape[-2:]
ref_im = resize(normalize(ref_im), args.image_size, scale_factor)
hB_, wB_ = ref_im.shape[-2:]
# create batch
batch = {}
batch['source_image'] = query_im.cuda()
batch['target_image'] = ref_im.cuda()
start = time.time()
matches, score, _ = matcher(batch,
pair_names_chunk = pair_names_split[args.chunk_idx]
pair_names_chunk = list(pair_names_chunk)
if args.skip_up_to != '':
pair_names_chunk = pair_names_chunk[pair_names_chunk.index(args.skip_up_to
) + 1:]
for pair in tqdm(pair_names_chunk):
src_fn = os.path.join(args.aachen_path, 'database_and_query',
'images_upright',
pair.split(' ')[0])
src_image = plt.imread(src_fn)
src = imreadth(src_fn)
hA, wA = src.shape[-2:]
src = resize(normalize(src), args.image_size, scale_factor)
hA_, wA_ = src.shape[-2:]
tgt_fn = os.path.join(args.aachen_path, 'database_and_query',
'images_upright',
pair.split(' ')[1])
tgt_image = plt.imread(tgt_fn)
tgt = imreadth(tgt_fn)
hB, wB = tgt.shape[-2:]
tgt = resize(normalize(tgt), args.image_size, scale_factor)
hB_, wB_ = tgt.shape[-2:]
with torch.no_grad():
result, scores, features = matcher(
{
'source_image': src,