2022-09-15 16:21:27 +02:00
|
|
|
import os
|
2022-04-15 18:11:51 +02:00
|
|
|
from pathlib import Path
|
|
|
|
|
|
|
|
import cv2
|
|
|
|
import pytest
|
2022-09-05 07:07:25 +02:00
|
|
|
import torch
|
2022-04-15 18:11:51 +02:00
|
|
|
|
|
|
|
from lama_cleaner.model_manager import ModelManager
|
2022-09-22 06:38:32 +02:00
|
|
|
from lama_cleaner.schema import Config, HDStrategy, LDMSampler, SDSampler
|
2022-04-15 18:11:51 +02:00
|
|
|
|
|
|
|
current_dir = Path(__file__).parent.absolute().resolve()
|
2023-03-25 15:46:28 +01:00
|
|
|
save_dir = current_dir / "result"
|
2022-09-30 15:39:23 +02:00
|
|
|
save_dir.mkdir(exist_ok=True, parents=True)
|
2023-03-25 15:46:28 +01:00
|
|
|
device = "cuda" if torch.cuda.is_available() else "cpu"
|
2022-11-14 11:19:50 +01:00
|
|
|
device = torch.device(device)
|
2022-04-15 18:11:51 +02:00
|
|
|
|
|
|
|
|
2023-03-25 15:46:28 +01:00
|
|
|
def get_data(
|
|
|
|
fx: float = 1,
|
|
|
|
fy: float = 1.0,
|
|
|
|
img_p=current_dir / "image.png",
|
|
|
|
mask_p=current_dir / "mask.png",
|
|
|
|
):
|
2022-09-15 16:21:27 +02:00
|
|
|
img = cv2.imread(str(img_p))
|
2022-04-15 18:11:51 +02:00
|
|
|
img = cv2.cvtColor(img, cv2.COLOR_BGRA2RGB)
|
2022-09-15 16:21:27 +02:00
|
|
|
mask = cv2.imread(str(mask_p), cv2.IMREAD_GRAYSCALE)
|
2022-10-21 04:36:55 +02:00
|
|
|
img = cv2.resize(img, None, fx=fx, fy=fy, interpolation=cv2.INTER_AREA)
|
|
|
|
mask = cv2.resize(mask, None, fx=fx, fy=fy, interpolation=cv2.INTER_NEAREST)
|
2022-04-15 18:11:51 +02:00
|
|
|
return img, mask
|
|
|
|
|
|
|
|
|
2022-07-14 10:49:03 +02:00
|
|
|
def get_config(strategy, **kwargs):
|
|
|
|
data = dict(
|
2022-04-15 18:11:51 +02:00
|
|
|
ldm_steps=1,
|
2022-07-14 10:49:03 +02:00
|
|
|
ldm_sampler=LDMSampler.plms,
|
2022-04-15 18:11:51 +02:00
|
|
|
hd_strategy=strategy,
|
|
|
|
hd_strategy_crop_margin=32,
|
|
|
|
hd_strategy_crop_trigger_size=200,
|
|
|
|
hd_strategy_resize_limit=200,
|
|
|
|
)
|
2022-07-14 10:49:03 +02:00
|
|
|
data.update(**kwargs)
|
|
|
|
return Config(**data)
|
2022-04-15 18:11:51 +02:00
|
|
|
|
|
|
|
|
2023-03-25 15:46:28 +01:00
|
|
|
def assert_equal(
|
|
|
|
model,
|
|
|
|
config,
|
|
|
|
gt_name,
|
|
|
|
fx: float = 1,
|
|
|
|
fy: float = 1,
|
|
|
|
img_p=current_dir / "image.png",
|
|
|
|
mask_p=current_dir / "mask.png",
|
|
|
|
):
|
2022-09-15 16:21:27 +02:00
|
|
|
img, mask = get_data(fx=fx, fy=fy, img_p=img_p, mask_p=mask_p)
|
2022-10-21 04:36:55 +02:00
|
|
|
print(f"Input image shape: {img.shape}")
|
2022-04-15 18:11:51 +02:00
|
|
|
res = model(img, mask, config)
|
2022-07-21 16:09:10 +02:00
|
|
|
cv2.imwrite(
|
2022-09-30 15:39:23 +02:00
|
|
|
str(save_dir / gt_name),
|
2022-07-21 16:09:10 +02:00
|
|
|
res,
|
|
|
|
[int(cv2.IMWRITE_JPEG_QUALITY), 100, int(cv2.IMWRITE_PNG_COMPRESSION), 0],
|
|
|
|
)
|
2022-04-15 18:11:51 +02:00
|
|
|
|
|
|
|
"""
|
|
|
|
Note that JPEG is lossy compression, so even if it is the highest quality 100,
|
2022-07-14 10:49:03 +02:00
|
|
|
when the saved images is reloaded, a difference occurs with the original pixel value.
|
|
|
|
If you want to save the original images as it is, save it as PNG or BMP.
|
2022-04-15 18:11:51 +02:00
|
|
|
"""
|
2022-07-14 10:49:03 +02:00
|
|
|
# gt = cv2.imread(str(current_dir / gt_name), cv2.IMREAD_UNCHANGED)
|
|
|
|
# assert np.array_equal(res, gt)
|
2022-04-15 18:11:51 +02:00
|
|
|
|
|
|
|
|
2023-03-25 15:46:28 +01:00
|
|
|
@pytest.mark.parametrize("strategy", [HDStrategy.ORIGINAL])
|
2022-08-22 17:24:02 +02:00
|
|
|
def test_mat(strategy):
|
2022-09-05 07:07:25 +02:00
|
|
|
model = ModelManager(name="mat", device=device)
|
2022-08-22 17:24:02 +02:00
|
|
|
cfg = get_config(strategy)
|
|
|
|
|
2023-03-25 15:46:28 +01:00
|
|
|
for _ in range(10):
|
|
|
|
assert_equal(
|
|
|
|
model, cfg, f"mat_{strategy.capitalize()}_result.png",
|
|
|
|
)
|
2022-09-02 04:37:30 +02:00
|
|
|
|