def main(args):
jsons = {
'train': 'dump/train_si284/deltafalse/data.json',
'dev': 'dump/test_dev93/deltafalse/data.json',
'test': 'dump/test_eval92/deltafalse/data.json'
}
spk2genders = {
'train': 'train_si284/spk2gender',
'dev': 'test_dev93/spk2gender',
'test': 'test_eval92/spk2gender'
}
utils.safe_copytree(args.data_root, args.temp_root)
if not os.path.exists(args.model_dir):
os.makedirs(args.model_dir)
logging.basicConfig(filename=os.path.join(args.model_dir, args.log_file),
filemode='a',
level=logging.INFO)
torch.manual_seed(args.seed)
torch.backends.cudnn.deterministic = args.determ
np.random.seed(args.seed)
if not args.eval_only:
train(args, jsons, spk2genders)
evaluate(args, jsons, spk2genders)
if args.cleanup:
utils.safe_rmtree(args.temp_root)
def main(args):
'''
Training and evaluation script for character-based CTC ASR on WSJ
dataset, pre-processed by ESPnet toolkit
'''
jsons = {
'train': 'dump/train_si284/deltafalse/data.json',
'dev': 'dump/test_dev93/deltafalse/data.json',
'test': 'dump/test_eval92/deltafalse/data.json'
}
# if the temporary directory contains a json, we'll assume it's correct
if not os.path.exists(os.path.join(args.temp_root, jsons['train'])):
# copy the data for faster reading than NFS
utils.safe_copytree(args.data_root, args.temp_root)
# if model_dir is specified, and it doesn't contain the log file, make it
log_file = os.path.join(args.model_dir, args.log_file)
if args.model_dir is not None and not os.path.exists(log_file):
utils.safe_makedirs(args.model_dir)
logging.basicConfig(filename=log_file, filemode='a', level=logging.INFO)
torch.manual_seed(args.seed)
torch.backends.cudnn.deterministic = not args.nondeterm
np.random.seed(args.seed)
if not args.eval_only:
utils.safe_json_dump(vars(args),
os.path.join(args.model_dir, 'args.json'))
epoch_stats = train(args, jsons)
utils.safe_json_dump(epoch_stats,
os.path.join(args.model_dir, 'epoch_stats.json'))
if args.eval_only:
data_root, temp_root = args.data_root, args.temp_root
test, cpu, seed, cleanup = args.test, args.cpu, args.seed, args.cleanup
with open(os.path.join(args.model_dir, 'args.json'), 'r') as f:
json_dict = json.load(f)
args = argparse.Namespace(**json_dict)
args.data_root, args.temp_root = data_root, temp_root
args.test, args.cpu, args.seed, args.cleanup = test, cpu, seed, cleanup
evaluate(args, jsons)
if args.cleanup:
utils.safe_rmtree(args.temp_root)
for file_path in files_2b_copied:
file_name = os.path.basename(file_path)
shutil.copyfile(file_path, os.path.join(eval_result_dirpath, file_name))
print("file '%s' is copyed into '%s'" % (file_name,eval_result_dirpath))
import sys,pathlib
from fp import pipe,cmap,cfilter
if __name__ == '__main__':
'''
python evaluator.py segnet.h5 imgs_dir output_dir
'''
segnet_model_path = sys.argv[1]
imgs_dir = sys.argv[2]
output_dir = sys.argv[3]
utils.safe_copytree(imgs_dir, output_dir,['*.*'])
segnet = model.unet(segnet_model_path, (None,None,1))
f = pipe(utils.file_paths,
cmap(lambda path: (cv2.imread(path,0), path)),
cfilter(lambda img_path: img_path[0] is not None),
cmap(lambda img_path: (utils.bgr_float32(img_path[0]), img_path[1]) ),
cmap(lambda im_p: (im_p[0].reshape((1,)+im_p[0].shape), im_p[1]) ),
cmap(lambda im_p: (inference(segnet,im_p[0]), im_p[1])))
old_parent_dir = pathlib.Path(imgs_dir).parts[-1]
for segmap_list, img_path in f(imgs_dir):
new_path = utils.make_dstpath(img_path, old_parent_dir, output_dir)
segmap = segmap_list[0]
segmap = (segmap.reshape(segmap.shape[:2]) * 255).astype(np.uint8)
create new argv[2] directory(tree structure preserved),
separate images in argv[1], and move rgb imgs to new directory.
ex)
python separator.py ./data/examples/ ./data/rgb
^~~~~~~~~~~~~~~~ ^~~~~~~~~~
origin img dir new directory for rgb imgs.
''')
is_grayscale = (lambda img: np.all(img[:, :, 0] == img[:, :, 1]) and np.all(
img[:, :, 1] == img[:, :, 2]))
if __name__ == '__main__':
mixed_imgs_path = sys.argv[1]
rgb_imgs_path = sys.argv[2]
utils.safe_copytree(mixed_imgs_path, rgb_imgs_path,
('*.jpg', '*.jpeg', '*.png'))
f = pipe(utils.file_paths, cmap(lambda path: (cv2.imread(path), path)),
cfilter(lambda img_path: img_path[0] is not None),
cfilter(lambda img_path: not is_grayscale(img_path[0])))
old_parent_dir = pathlib.Path(mixed_imgs_path).parts[-1]
timer = utils.ElapsedTimer('moving in')
for img, img_path in f(mixed_imgs_path):
new_path = utils.make_dstpath(img_path, old_parent_dir, rgb_imgs_path)
#print(img_path, old_parent_dir, new_path)
shutil.move(img_path, new_path)
timer.elapsed_time()
#print(y,x)
#cv2.imshow('img',img[y:y+h, x:x+w]); cv2.waitKey(0)
yield '%s_%d_%d.png' % (img_name,y,x), img[y:y+h, x:x+w]
'''
if __name__ == '__main__':
def path2path_img(path):
return (path, cv2.imread(path))
crop_size = int(sys.argv[1])
imgs_dir = sys.argv[2]
pieces_dir = sys.argv[3]
utils.safe_copytree(imgs_dir, pieces_dir, ['*.jpg', '*.jpeg', '*.png'])
timer = utils.ElapsedTimer('Total Cutting')
#-------------------------------------------------------------
pieces \
= pipe(utils.file_paths,
cmap(path2path_img),
cfilter(lambda path_img:path_img[1] is not None),
cmap(lambda pair: path_img2path_pieces(pair,crop_size,imgs_dir,pieces_dir)),
flatten)(imgs_dir)
for path, img in pieces:
#print(path)
#print(path)
#cv2.imwrite(path, img)
cv2.imwrite(path, img)