Init object remover
Browse files
FGT_codes
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
Subproject commit f7bc8a2c520ef862d0a4d28a38334604eb41410c
|
SiamMask
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
Subproject commit 0eaac33050fdcda81c9a25aa307fffa74c182e36
|
app.py
ADDED
@@ -0,0 +1,259 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from PIL import Image
|
2 |
+
import gradio as gr
|
3 |
+
from video_inpainting import video_inpainting
|
4 |
+
from tools.test import *
|
5 |
+
from custom import Custom
|
6 |
+
from types import SimpleNamespace
|
7 |
+
import torch
|
8 |
+
import numpy as np
|
9 |
+
import torchvision
|
10 |
+
import cv2
|
11 |
+
import sys
|
12 |
+
from os.path import exists, join, basename, splitext
|
13 |
+
import os
|
14 |
+
project_name = './video-object-remover'
|
15 |
+
|
16 |
+
sys.path.append(project_name)
|
17 |
+
|
18 |
+
sys.path.append(join(project_name, 'SiamMask',
|
19 |
+
'experiments', 'siammask_sharp'))
|
20 |
+
sys.path.append(join(project_name, 'SiamMask', 'models'))
|
21 |
+
sys.path.append(join(project_name, 'SiamMask'))
|
22 |
+
|
23 |
+
exp_path = join(project_name, 'SiamMask/experiments/siammask_sharp')
|
24 |
+
pretrained_path1 = join(exp_path, 'SiamMask_DAVIS.pth')
|
25 |
+
|
26 |
+
sys.path.append(join(project_name, 'FGT_codes'))
|
27 |
+
sys.path.append(join(project_name, 'FGT_codes', 'tool'))
|
28 |
+
sys.path.append(join(project_name, 'FGT_codes', 'LAFC', 'flowCheckPoint'))
|
29 |
+
sys.path.append(join(project_name, 'FGT_codes', 'LAFC', 'checkpoint'))
|
30 |
+
sys.path.append(join(project_name, 'FGT_codes', 'FGT', 'checkpoint'))
|
31 |
+
sys.path.append(join(project_name, 'FGT_codes', 'LAFC',
|
32 |
+
'flowCheckPoint', 'raft-things.pth'))
|
33 |
+
|
34 |
+
torch.set_grad_enabled(False)
|
35 |
+
|
36 |
+
# init SiamMask
|
37 |
+
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
38 |
+
cfg = load_config(SimpleNamespace(config=join(exp_path, 'config_davis.json')))
|
39 |
+
siammask = Custom(anchors=cfg['anchors'])
|
40 |
+
siammask = load_pretrain(siammask, pretrained_path1)
|
41 |
+
siammask = siammask.eval().to(device)
|
42 |
+
|
43 |
+
# constants
|
44 |
+
object_x = 0
|
45 |
+
object_y = 0
|
46 |
+
object_width = 0
|
47 |
+
object_height = 0
|
48 |
+
original_frame_list = []
|
49 |
+
mask_list = []
|
50 |
+
|
51 |
+
parser = argparse.ArgumentParser()
|
52 |
+
parser.add_argument('--opt', default='configs/object_removal.yaml',
|
53 |
+
help='Please select your config file for inference')
|
54 |
+
# video completion
|
55 |
+
parser.add_argument('--mode', default='object_removal', choices=[
|
56 |
+
'object_removal', 'watermark_removal', 'video_extrapolation'], help="modes: object_removal / video_extrapolation")
|
57 |
+
parser.add_argument(
|
58 |
+
'--path', default='/myData/davis_resized/walking', help="dataset for evaluation")
|
59 |
+
parser.add_argument(
|
60 |
+
'--path_mask', default='/myData/dilateAnnotations_4/walking', help="mask for object removal")
|
61 |
+
parser.add_argument(
|
62 |
+
'--outroot', default='quick_start/walking3', help="output directory")
|
63 |
+
parser.add_argument('--consistencyThres', dest='consistencyThres', default=5, type=float,
|
64 |
+
help='flow consistency error threshold')
|
65 |
+
parser.add_argument('--alpha', dest='alpha', default=0.1, type=float)
|
66 |
+
parser.add_argument('--Nonlocal', dest='Nonlocal',
|
67 |
+
default=False, type=bool)
|
68 |
+
|
69 |
+
# RAFT
|
70 |
+
parser.add_argument(
|
71 |
+
'--raft_model', default='../LAFC/flowCheckPoint/raft-things.pth', help="restore checkpoint")
|
72 |
+
parser.add_argument('--small', action='store_true', help='use small model')
|
73 |
+
parser.add_argument('--mixed_precision',
|
74 |
+
action='store_true', help='use mixed precision')
|
75 |
+
parser.add_argument('--alternate_corr', action='store_true',
|
76 |
+
help='use efficent correlation implementation')
|
77 |
+
|
78 |
+
# LAFC
|
79 |
+
parser.add_argument('--lafc_ckpts', type=str, default='../LAFC/checkpoint')
|
80 |
+
|
81 |
+
# FGT
|
82 |
+
parser.add_argument('--fgt_ckpts', type=str, default='../FGT/checkpoint')
|
83 |
+
|
84 |
+
# extrapolation
|
85 |
+
parser.add_argument('--H_scale', dest='H_scale', default=2,
|
86 |
+
type=float, help='H extrapolation scale')
|
87 |
+
parser.add_argument('--W_scale', dest='W_scale', default=2,
|
88 |
+
type=float, help='W extrapolation scale')
|
89 |
+
|
90 |
+
# Image basic information
|
91 |
+
parser.add_argument('--imgH', type=int, default=256)
|
92 |
+
parser.add_argument('--imgW', type=int, default=432)
|
93 |
+
parser.add_argument('--flow_mask_dilates', type=int, default=8)
|
94 |
+
parser.add_argument('--frame_dilates', type=int, default=0)
|
95 |
+
|
96 |
+
parser.add_argument('--gpu', type=int, default=0)
|
97 |
+
|
98 |
+
# FGT inference parameters
|
99 |
+
parser.add_argument('--step', type=int, default=10)
|
100 |
+
parser.add_argument('--num_ref', type=int, default=-1)
|
101 |
+
parser.add_argument('--neighbor_stride', type=int, default=5)
|
102 |
+
|
103 |
+
# visualization
|
104 |
+
parser.add_argument('--vis_flows', action='store_true',
|
105 |
+
help='Visualize the initialized flows')
|
106 |
+
parser.add_argument('--vis_completed_flows',
|
107 |
+
action='store_true', help='Visualize the completed flows')
|
108 |
+
parser.add_argument('--vis_prop', action='store_true',
|
109 |
+
help='Visualize the frames after stage-I filling (flow guided content propagation)')
|
110 |
+
parser.add_argument('--vis_frame', action='store_true',
|
111 |
+
help='Visualize frames')
|
112 |
+
|
113 |
+
args = parser.parse_args()
|
114 |
+
|
115 |
+
|
116 |
+
def getBoundaries(mask):
|
117 |
+
if mask is None:
|
118 |
+
return 0, 0, 0, 0
|
119 |
+
|
120 |
+
indexes = np.where((mask == [255, 255, 255]).all(axis=2))
|
121 |
+
print(indexes)
|
122 |
+
x1 = min(indexes[1])
|
123 |
+
y1 = min(indexes[0])
|
124 |
+
x2 = max(indexes[1])
|
125 |
+
y2 = max(indexes[0])
|
126 |
+
|
127 |
+
return x1, y1, (x2-x1), (y2-y1)
|
128 |
+
|
129 |
+
|
130 |
+
def track_and_mask(vid, original_frame, masked_frame):
|
131 |
+
x, y, w, h = getBoundaries(masked_frame)
|
132 |
+
f = 0
|
133 |
+
|
134 |
+
video_capture = cv2.VideoCapture()
|
135 |
+
if video_capture.open(vid):
|
136 |
+
width, height = int(video_capture.get(cv2.CAP_PROP_FRAME_WIDTH)), int(
|
137 |
+
video_capture.get(cv2.CAP_PROP_FRAME_HEIGHT))
|
138 |
+
fps = video_capture.get(cv2.CAP_PROP_FPS)
|
139 |
+
|
140 |
+
# can't write out mp4, so try to write into an AVI file
|
141 |
+
video_writer = cv2.VideoWriter(
|
142 |
+
"output.avi", cv2.VideoWriter_fourcc(*'MP42'), fps, (width, height))
|
143 |
+
video_writer2 = cv2.VideoWriter(
|
144 |
+
"output_mask.avi", cv2.VideoWriter_fourcc(*'MP42'), fps, (width, height))
|
145 |
+
|
146 |
+
while video_capture.isOpened():
|
147 |
+
ret, frame = video_capture.read()
|
148 |
+
|
149 |
+
if not ret:
|
150 |
+
break
|
151 |
+
|
152 |
+
# frame = cv2.resize(frame, (w - w % 8, h - h % 8))
|
153 |
+
if f == 0:
|
154 |
+
target_pos = np.array([x + w / 2, y + h / 2])
|
155 |
+
target_sz = np.array([w, h])
|
156 |
+
# init tracker
|
157 |
+
state = siamese_init(
|
158 |
+
frame, target_pos, target_sz, siammask, cfg['hp'], device=device)
|
159 |
+
else:
|
160 |
+
# track
|
161 |
+
state = siamese_track(
|
162 |
+
state, frame, mask_enable=True, refine_enable=True, device=device)
|
163 |
+
location = state['ploygon'].flatten()
|
164 |
+
mask = state['mask'] > state['p'].seg_thr
|
165 |
+
frame[:, :, 2] = (mask > 0) * 255 + \
|
166 |
+
(mask == 0) * frame[:, :, 2]
|
167 |
+
|
168 |
+
mask = mask.astype(np.uint8) # convert to an unsigned byte
|
169 |
+
mask = mask * 255
|
170 |
+
mask_list.append(mask)
|
171 |
+
cv2.polylines(frame, [np.int0(location).reshape(
|
172 |
+
(-1, 1, 2))], True, (0, 255, 0), 3)
|
173 |
+
|
174 |
+
original_frame_list.append(frame)
|
175 |
+
mask_list.append(mask)
|
176 |
+
|
177 |
+
video_writer.write(frame)
|
178 |
+
video_writer2.write(mask)
|
179 |
+
f = f + 1
|
180 |
+
|
181 |
+
video_capture.release()
|
182 |
+
video_writer.release()
|
183 |
+
video_writer2.release()
|
184 |
+
|
185 |
+
else:
|
186 |
+
print("can't open the given input video file!")
|
187 |
+
|
188 |
+
return "output.mp4"
|
189 |
+
|
190 |
+
|
191 |
+
def inpaint_video():
|
192 |
+
video_inpainting(args, original_frame_list, mask_list)
|
193 |
+
|
194 |
+
return "output.mp4"
|
195 |
+
|
196 |
+
|
197 |
+
def get_first_frame(video):
|
198 |
+
video_capture = cv2.VideoCapture()
|
199 |
+
if video_capture.open(video):
|
200 |
+
width, height = int(video_capture.get(cv2.CAP_PROP_FRAME_WIDTH)), int(
|
201 |
+
video_capture.get(cv2.CAP_PROP_FRAME_HEIGHT))
|
202 |
+
|
203 |
+
if video_capture.isOpened():
|
204 |
+
ret, frame = video_capture.read()
|
205 |
+
|
206 |
+
RGB_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
|
207 |
+
return RGB_frame
|
208 |
+
|
209 |
+
|
210 |
+
def drawRectangle(frame, mask):
|
211 |
+
x1, y1, x2, y2 = getBoundaries(mask)
|
212 |
+
|
213 |
+
return cv2.rectangle(frame, (int(x1), int(y1)), (int(x2), int(y2)), (0, 255, 0), 2)
|
214 |
+
|
215 |
+
|
216 |
+
def getStartEndPoints(mask):
|
217 |
+
if mask is None:
|
218 |
+
return 0, 0, 0, 0
|
219 |
+
|
220 |
+
indexes = np.where((mask == [255, 255, 255]).all(axis=2))
|
221 |
+
print(indexes)
|
222 |
+
x1 = min(indexes[1])
|
223 |
+
y1 = min(indexes[0])
|
224 |
+
x2 = max(indexes[1])
|
225 |
+
y2 = max(indexes[0])
|
226 |
+
|
227 |
+
return x1, y1, x2, y2
|
228 |
+
|
229 |
+
|
230 |
+
with gr.Blocks() as demo:
|
231 |
+
with gr.Row():
|
232 |
+
with gr.Column(scale=2):
|
233 |
+
with gr.Row():
|
234 |
+
in_video = gr.Video()
|
235 |
+
with gr.Row():
|
236 |
+
first_frame = gr.ImageMask()
|
237 |
+
with gr.Row():
|
238 |
+
approve_mask = gr.Button(value="Approve Mask")
|
239 |
+
with gr.Column(scale=1):
|
240 |
+
with gr.Row():
|
241 |
+
original_image = gr.Image(interactive=False)
|
242 |
+
with gr.Row():
|
243 |
+
masked_image = gr.Image(interactive=False)
|
244 |
+
with gr.Column(scale=2):
|
245 |
+
out_video = gr.Video()
|
246 |
+
out_video_inpaint = gr.Video()
|
247 |
+
track_mask = gr.Button(value="Track and Mask")
|
248 |
+
inpaint = gr.Button(value="Inpaint")
|
249 |
+
|
250 |
+
in_video.change(fn=get_first_frame, inputs=[
|
251 |
+
in_video], outputs=[first_frame])
|
252 |
+
approve_mask.click(lambda x: [x['image'], x['mask']], first_frame, [
|
253 |
+
original_image, masked_image])
|
254 |
+
track_mask.click(fn=track_and_mask, inputs=[
|
255 |
+
in_video, original_image, masked_image], outputs=[out_video])
|
256 |
+
inpaint.click(fn=inpaint_video, outputs=[out_video_inpaint])
|
257 |
+
|
258 |
+
|
259 |
+
demo.launch(share=True, debug=True)
|
requirements.txt
ADDED
@@ -0,0 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
torch==1.10.1
|
2 |
+
torchvision==0.11.2
|
3 |
+
cvbase==0.5.5
|
4 |
+
imageio==2.6.1
|
5 |
+
matplotlib==3.1.1
|
6 |
+
numpy==1.22.2
|
7 |
+
opencv-python
|
8 |
+
Pillow
|
9 |
+
PyYAML
|
10 |
+
scikit-image
|
11 |
+
scipy
|
12 |
+
tensorboardX
|
13 |
+
imageio-ffmpeg
|
14 |
+
Cython==0.29.34
|
15 |
+
colorama==0.3.9
|
16 |
+
requests==2.21.0
|
17 |
+
fire==0.1.3
|
18 |
+
numba==0.39.0
|
19 |
+
h5py==2.8.0
|
20 |
+
tqdm==4.29.1
|