Spaces:
Runtime error
Runtime error
Update demo_gradio.py
Browse files- demo_gradio.py +21 -28
demo_gradio.py
CHANGED
|
@@ -6,10 +6,8 @@
|
|
| 6 |
import torch
|
| 7 |
import torch.nn.functional as F
|
| 8 |
import os
|
| 9 |
-
from runpy import run_path
|
| 10 |
from skimage import img_as_ubyte
|
| 11 |
import cv2
|
| 12 |
-
from tqdm import tqdm
|
| 13 |
import argparse
|
| 14 |
|
| 15 |
parser = argparse.ArgumentParser(description='Test Restormer on your own images')
|
|
@@ -25,45 +23,36 @@ parser.add_argument('--task', required=True, type=str, help='Task to run', choic
|
|
| 25 |
args = parser.parse_args()
|
| 26 |
|
| 27 |
|
| 28 |
-
def get_weights_and_parameters(task, parameters):
|
| 29 |
-
if task == 'Motion_Deblurring':
|
| 30 |
-
weights = os.path.join('Motion_Deblurring', 'pretrained_models', 'motion_deblurring.pth')
|
| 31 |
-
elif task == 'Single_Image_Defocus_Deblurring':
|
| 32 |
-
weights = os.path.join('Defocus_Deblurring', 'pretrained_models', 'single_image_defocus_deblurring.pth')
|
| 33 |
-
elif task == 'Deraining':
|
| 34 |
-
weights = os.path.join('Deraining', 'pretrained_models', 'deraining.pth')
|
| 35 |
-
elif task == 'Real_Denoising':
|
| 36 |
-
weights = os.path.join('Denoising', 'pretrained_models', 'real_denoising.pth')
|
| 37 |
-
parameters['LayerNorm_type'] = 'BiasFree'
|
| 38 |
-
return weights, parameters
|
| 39 |
-
|
| 40 |
task = args.task
|
| 41 |
out_dir = os.path.join(args.result_dir, task)
|
| 42 |
|
| 43 |
os.makedirs(out_dir, exist_ok=True)
|
| 44 |
|
| 45 |
-
# Get model weights and parameters
|
| 46 |
-
parameters = {'inp_channels':3, 'out_channels':3, 'dim':48, 'num_blocks':[4,6,6,8], 'num_refinement_blocks':4, 'heads':[1,2,4,8], 'ffn_expansion_factor':2.66, 'bias':False, 'LayerNorm_type':'WithBias', 'dual_pixel_task':False}
|
| 47 |
-
weights, parameters = get_weights_and_parameters(task, parameters)
|
| 48 |
-
|
| 49 |
-
load_arch = run_path(os.path.join('basicsr', 'models', 'archs', 'restormer_arch.py'))
|
| 50 |
-
model = load_arch['Restormer'](**parameters)
|
| 51 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 52 |
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
| 53 |
# device = torch.device('cpu')
|
|
|
|
| 54 |
|
| 55 |
model = model.to(device)
|
| 56 |
-
checkpoint = torch.load(weights)
|
| 57 |
-
model.load_state_dict(checkpoint['params'])
|
| 58 |
-
|
| 59 |
-
model.eval()
|
| 60 |
|
| 61 |
|
| 62 |
img_multiple_of = 8
|
| 63 |
|
| 64 |
|
| 65 |
with torch.inference_mode():
|
| 66 |
-
|
|
|
|
|
|
|
|
|
|
| 67 |
img = cv2.cvtColor(cv2.imread(args.input_path), cv2.COLOR_BGR2RGB)
|
| 68 |
|
| 69 |
input_ = torch.from_numpy(img).float().div(255.).permute(2,0,1).unsqueeze(0).to(device)
|
|
@@ -74,10 +63,14 @@ with torch.inference_mode():
|
|
| 74 |
padh = H-h if h%img_multiple_of!=0 else 0
|
| 75 |
padw = W-w if w%img_multiple_of!=0 else 0
|
| 76 |
input_ = F.pad(input_, (0,padw,0,padh), 'reflect')
|
| 77 |
-
|
|
|
|
| 78 |
restored = torch.clamp(model(input_),0,1)
|
| 79 |
-
|
| 80 |
# Unpad the output
|
| 81 |
restored = img_as_ubyte(restored[:,:,:h,:w].permute(0, 2, 3, 1).cpu().detach().numpy()[0])
|
| 82 |
|
| 83 |
-
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
import torch
|
| 7 |
import torch.nn.functional as F
|
| 8 |
import os
|
|
|
|
| 9 |
from skimage import img_as_ubyte
|
| 10 |
import cv2
|
|
|
|
| 11 |
import argparse
|
| 12 |
|
| 13 |
parser = argparse.ArgumentParser(description='Test Restormer on your own images')
|
|
|
|
| 23 |
args = parser.parse_args()
|
| 24 |
|
| 25 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
task = args.task
|
| 27 |
out_dir = os.path.join(args.result_dir, task)
|
| 28 |
|
| 29 |
os.makedirs(out_dir, exist_ok=True)
|
| 30 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 31 |
|
| 32 |
+
if task == 'Motion_Deblurring':
|
| 33 |
+
model = torch.jit.load('motion_deblurring.pt')
|
| 34 |
+
elif task == 'Single_Image_Defocus_Deblurring':
|
| 35 |
+
model = torch.jit.load('single_image_defocus_deblurring.pt')
|
| 36 |
+
elif task == 'Deraining':
|
| 37 |
+
model = torch.jit.load('deraining.pt')
|
| 38 |
+
elif task == 'Real_Denoising':
|
| 39 |
+
model = torch.jit.load('real_denoising.pt')
|
| 40 |
+
|
| 41 |
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
| 42 |
# device = torch.device('cpu')
|
| 43 |
+
# stx()
|
| 44 |
|
| 45 |
model = model.to(device)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 46 |
|
| 47 |
|
| 48 |
img_multiple_of = 8
|
| 49 |
|
| 50 |
|
| 51 |
with torch.inference_mode():
|
| 52 |
+
if torch.cuda.is_available():
|
| 53 |
+
torch.cuda.ipc_collect()
|
| 54 |
+
torch.cuda.empty_cache()
|
| 55 |
+
|
| 56 |
img = cv2.cvtColor(cv2.imread(args.input_path), cv2.COLOR_BGR2RGB)
|
| 57 |
|
| 58 |
input_ = torch.from_numpy(img).float().div(255.).permute(2,0,1).unsqueeze(0).to(device)
|
|
|
|
| 63 |
padh = H-h if h%img_multiple_of!=0 else 0
|
| 64 |
padw = W-w if w%img_multiple_of!=0 else 0
|
| 65 |
input_ = F.pad(input_, (0,padw,0,padh), 'reflect')
|
| 66 |
+
|
| 67 |
+
# print(h,w)
|
| 68 |
restored = torch.clamp(model(input_),0,1)
|
| 69 |
+
|
| 70 |
# Unpad the output
|
| 71 |
restored = img_as_ubyte(restored[:,:,:h,:w].permute(0, 2, 3, 1).cpu().detach().numpy()[0])
|
| 72 |
|
| 73 |
+
out_path = os.path.join(out_dir, os.path.split(args.input_path)[-1])
|
| 74 |
+
cv2.imwrite(out_path,cv2.cvtColor(restored, cv2.COLOR_RGB2BGR))
|
| 75 |
+
|
| 76 |
+
# print(f"\nRestored images are saved at {out_dir}")
|