-
Notifications
You must be signed in to change notification settings - Fork 8
/
inference.py
189 lines (160 loc) · 7.3 KB
/
inference.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
import torch
import torch.nn as nn
import torch.optim as optim
import config
import os
import cv2
from typing import Generator, List, Tuple
from generator_model import Generator
import argparse
from utils import load_checkpoint
from more_itertools import chunked
from tqdm import tqdm
import numpy as np
import albumentations as A
from albumentations.pytorch import ToTensorV2
from surface_extractor import GuidedFilter
def parser():
parser = argparse.ArgumentParser(description="inference.py: Model inference script of White-box Cartoonization.")
parser.add_argument('-s',"--source",required=True,
help="filepath to a source image or a video or a images folder.")
parser.add_argument('-w',"--weight_path", required=True,
help="path to model weight file.")
parser.add_argument("--batch_size", type=int, default= config.BATCH_SIZE,
help="batch size for video inference. default size:"+f"{config.BATCH_SIZE}")
parser.add_argument("--dest_folder", type=str, required=True,
help="Destination folder path for saving results.")
parser.add_argument("--suffix", type=str, default= "_infered",
help="Output suffix.")
return parser.parse_args()
def check_arguments_errors(args):
if(check_format(args.source) == None):
raise(ValueError("Invalid input path {}".format(os.path.abspath(args.source))))
if not os.path.isfile(args.weight_path):
raise(ValueError("Invalid model weight path {}".format(os.path.abspath(args.weight_path))))
def check_format(source_path):
if(os.path.isdir(source_path)):
return 'folder'
IMG_FORMATS = 'bmp', 'dng', 'jpeg', 'jpg', 'mpo', 'png', 'tif', 'tiff', 'webp' # include image suffixes
VID_FORMATS = 'asf', 'avi', 'gif', 'm4v', 'mkv', 'mov', 'mp4', 'mpeg', 'mpg', 'ts', 'wmv' # include video suffixes
if(source_path.split(".")[-1] in IMG_FORMATS):
return 'image'
if(source_path.split(".")[-1] in VID_FORMATS):
return 'video'
return None
def get_scaled_dim(height, width):
height, width = reduce_to_scale([height, width], [config.IMAGE_SIZE, config.IMAGE_SIZE], scale=32)
return height, width
def reduce_to_scale(img_hw: List[int], min_hw: List[int], scale: int) -> Tuple[int]:
im_h, im_w = img_hw
if im_h <= min_hw[0]:
im_h = min_hw[0]
else:
x = im_h % scale
im_h = im_h - x
if im_w < min_hw[1]:
im_w = min_hw[1]
else:
y = im_w % scale
im_w = im_w - y
return (im_h, im_w)
def get_transform(height, width):
height, width = get_scaled_dim(height=height, width=width)
transform_inference = A.Compose(
[
A.Resize(width=width, height=height),
A.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5], max_pixel_value=255),
ToTensorV2(),
]
)
return transform_inference
#===============================================
#NOTE Equation 9 in the paper, adjust sharpness of output
# If no post_processing, noise will be found
def post_processing(x, G_x, r: int=1):
extract_surface = GuidedFilter()
return extract_surface.process(x, G_x, r=r)
#===============================================
def infer_batch(img, model):
img = img.to(config.DEVICE)
out = model.forward(img)
out = torch.tanh(out)
out = post_processing(img, out)
#TODO fix this later
unnormalized_out = out*0.5+0.5
bgr_img = (unnormalized_out.permute((0,2,3,1)).detach().to('cpu').numpy()*255).astype('uint8')
return bgr_img
def infer_one_image(source_path, output_path, model, suffix):
rgb_image = cv2.cvtColor(cv2.imread(source_path, flags=cv2.IMREAD_COLOR), cv2.COLOR_BGR2RGB)
height = rgb_image.shape[0]
width = rgb_image.shape[1]
transform_inference = get_transform(height=height, width=width)
norm_image = transform_inference(image=rgb_image)["image"]
out_img = infer_batch(norm_image[None, ...], model)
out_img = out_img[0]
cv2.imwrite(output_path+suffix+'.png', cv2.cvtColor(out_img, cv2.COLOR_RGB2BGR))
#reference: https://github.com/zhen8838/AnimeStylized/blob/main/utils/video.py
def get_video_reader(path):
video_cap = cv2.VideoCapture(path)
length = int(video_cap.get(cv2.CAP_PROP_FRAME_COUNT))
fps = int(video_cap.get(cv2.CAP_PROP_FPS))
height = int(video_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
width = int(video_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
def gen(cap):
while True:
ret, frame = cap.read()
if ret:
yield frame
else:
cap.release()
break
return gen(video_cap), length, fps, height, width
def get_video_writer(path, fps, height, width):
video_writer = cv2.VideoWriter(filename=path, fourcc=cv2.VideoWriter_fourcc(*'mp4v'), fps=fps, frameSize=(width, height))
return video_writer
def infer_video(source_path, output_path, batch_size, model, suffix):
video_read, length, fps, height, width = get_video_reader(source_path)
#update scaled height and width
height, width = get_scaled_dim(height=height, width=width)
transform_inference = get_transform(height=height, width=width)
output_path = output_path+suffix+'.mp4'
video_write = get_video_writer(output_path, fps, height, width)
for frames in tqdm(chunked(video_read, batch_size), total=length // batch_size):
norm_imgs = torch.stack(
[transform_inference(image=cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))["image"] for frame in frames]
)
output_imgs = infer_batch(norm_imgs, model)
for img in output_imgs:
video_write.write(cv2.cvtColor(img, cv2.COLOR_RGB2BGR))
video_write.release()
def infer_fn(args):
print("Using Device: " + config.DEVICE)
gen = Generator(img_channels=3).to(config.DEVICE)
opt_gen = optim.Adam(gen.parameters(), lr=config.LEARNING_RATE, betas=(0.5, 0.999))
load_checkpoint(gen, opt_gen, config.LEARNING_RATE, path=args.weight_path)
gen.eval()
if(check_format(args.source) == 'video'):
infer_video(args.source, args.source, batch_size=args.batch_size, model=gen, suffix=args.suffix)
elif(check_format(args.source) == 'image'):
infer_one_image(args.source, args.source, model=gen, suffix=args.suffix)
if(check_format(args.source) == 'folder'):
#Create Inference folder
dest_folder = args.dest_folder
if not os.path.exists(dest_folder):
os.makedirs(dest_folder)
list_files = os.listdir(args.source)
print(f'{len(list_files)} files found in the folder {args.source}. Output folder: {dest_folder}')
for file in list_files:
input_path = os.path.join(args.source,file)
output_path = os.path.join(dest_folder,file)
if(check_format(input_path) == 'video'):
infer_video(input_path, output_path, batch_size=args.batch_size, model=gen, suffix=args.suffix)
print(f"Finished inferencing file: {file}")
elif(check_format(input_path) == 'image'):
infer_one_image(input_path, output_path, model=gen, suffix=args.suffix)
print(f"Finished inferencing file: {file}")
print("=> Finish Inference.")
if __name__ == "__main__":
args = parser()
check_arguments_errors(args)
infer_fn(args)