forked from Jyxarthur/flowsam
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathseq_level_postprocess.py
180 lines (149 loc) · 8.52 KB
/
seq_level_postprocess.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
import os
import cv2
from argparse import ArgumentParser
from scipy.optimize import linear_sum_assignment
from utils import imwrite_indexed
def save_indexed(filename, img):
""" Save image with given colour palette """
c,h,w = np.shape(img)
output_mask = np.zeros_like(img[0])
for score_idx in range(c):
output_mask = output_mask * (1 - img[score_idx]) + img[score_idx] * (score_idx + 1)
img = np.clip(output_mask, 0, c + 1).astype(np.uint8)
img = (5+img) * (img > 0)
color_palette = np.array([[0,0,0],[128, 0, 0], [0, 128, 0], [128, 128, 0], [0, 0, 128], [128, 0, 128], [0, 128, 128], [128, 128, 128], [64, 0, 0], [64, 128, 0], [191, 0, 0], [0, 64, 0], [0, 191, 0], [0, 0, 64], [0, 0, 191], [0, 64, 128], [128, 0, 64], [128, 64, 0], [0, 128, 64], [64, 0, 128], [191, 191, 0], [0, 191, 191]]).astype(np.uint8)
imwrite_indexed(filename, img, color_palette)
def threeway_hungarian(m1, m2, m3, thres = 0.5, emp=False):
m1 = (m1>thres)
m2 = (m2>thres)
m3 = (m3>thres)
ious_23 = iou(m2[:, None], m3[None], emp=False)
orig_idx, hung_idx = linear_sum_assignment(-ious_23)
m3_aligned = m3[hung_idx]
i23 = ious_23[orig_idx, hung_idx]
ious_13 = iou(m1[:, None], m3_aligned[None], emp=False)
ious_12 = iou(m1[:, None], m2[None], emp=False)
orig_idx_13, hung_idx_13 = linear_sum_assignment(-ious_13)
orig_idx_12, hung_idx_12 = linear_sum_assignment(-ious_12)
return hung_idx_12, (hung_idx_12 == hung_idx_13) *1.
def multiway_hungarian(p, m, b1, b2, f1, f2):
hung_idx, s1 = threeway_hungarian(p, m, b1)
_, s2 = threeway_hungarian(p, m, b2)
_, s3 = threeway_hungarian(p, m, f1)
_, s4 = threeway_hungarian(p, m, f2)
m_aligned = m[hung_idx]
out = []
s = s1+s2+s3+s4
for i in range(s.shape[0]):
if s[i] >= 3: #use 4 for flow-only
out.append(m_aligned[i])
elif s[i] == 0:
out.append(p[i])
else:
out.append(m_aligned[i]+p[i])
out = (np.stack(out,0) > 0.5) * 1.
return out
def main(args):
save_mask = True
orig_iouss = []
seq_ious = []
for cat in args.cats:
image_dir = os.path.join(args.rgb_dir, cat)
obj_num = args.obj_nums[cat]
orig_ious = []
preds = []
gts = []
masks = []
last_pred = []
for idx, image_path in enumerate(sorted(os.listdir(image_dir))):
m = processMultiSeg(os.path.join(args.pred_dir, 'modal', cat, image_path.replace(".jpg", ".png")))
f = processMultiSeg(os.path.join(args.flow_pred_dir, 'modal', cat, image_path.replace(".jpg", ".png")))[:5]
masks.append(remove_overlapping_masks(np.concatenate([m,f],0)))
gt = processMultiSeg(os.path.join(args.gt_dir, cat, image_path.replace(".jpg", ".png")))[:obj_num]
gts.append(gt)
for idx, image_path in enumerate(sorted(os.listdir(image_dir))):
mask = masks[idx]
if idx > 0:
flow = read(os.path.join(args.flow_b1_dir, cat, image_path.replace('jpg','flo')))
prev_mask = np.stack([warp_flow(last_pred[-1][i], flow) for i in range(last_pred[-1].shape[0])], 0)
b1_mask = masks[idx-1]
b1_mask = np.stack([warp_flow(b1_mask[i], flow) for i in range(b1_mask.shape[0])], 0)
else:
prev_mask = mask
b1_mask = mask
if idx > 1:
flow = read(os.path.join(args.flow_b2_dir, cat, image_path.replace('jpg','flo')))
b2_mask = masks[idx-2]
b2_mask = np.stack([warp_flow(b2_mask[i], flow) for i in range(b2_mask.shape[0])], 0)
last_pred = last_pred[-2:]
else:
b2_mask = mask
if idx < len(masks) - 1:
flow = read(os.path.join(args.flow_f1_dir, cat, image_path.replace('jpg','flo')))
f1_mask = masks[idx+1]
f1_mask = np.stack([warp_flow(f1_mask[i], flow) for i in range(f1_mask.shape[0])], 0)
else:
f1_mask = mask
if idx < len(masks) - 2:
flow = read(os.path.join(args.flow_f2_dir, cat, image_path.replace('jpg','flo')))
f2_mask = masks[idx+2]
f2_mask = np.stack([warp_flow(f2_mask[i], flow) for i in range(f2_mask.shape[0])], 0)
else:
f2_mask = mask
output_masks = multiway_hungarian(prev_mask, mask, b1_mask, b2_mask, f1_mask, f2_mask)
last_pred.append(output_masks)
output_masks = remove_overlapping_masks(output_masks)
preds.append(output_masks)
orig_iou, _ = hungarian_iou(mask, gts[idx], emp=True)
orig_ious.append(orig_iou)
if save_mask:
save_dir = "{}".format(args.save_dir)
os.makedirs(save_dir, exist_ok=True)
save_dir2 = "{}/{}_seq/".format(args.save_dir,args.dataset)
os.makedirs(save_dir2, exist_ok=True)
os.makedirs(os.path.join(save_dir2,cat), exist_ok=True)
save_indexed(os.path.join(save_dir2, cat, os.path.basename(image_path).replace(".jpg", ".png")), output_masks.astype(np.uint8))
save_dir2 = "{}/{}_seq_hung/".format(args.save_dir,args.dataset)
os.makedirs(save_dir2, exist_ok=True)
os.makedirs(os.path.join(save_dir2,cat), exist_ok=True)
save_indexed(os.path.join(save_dir2, cat, os.path.basename(image_path).replace(".jpg", ".png")), hungarian_iou(output_masks, gts[idx], emp=True)[1].astype(np.uint8))
seq_iou, preds = seq_hungarian_iou(preds, gts)
orig_ious = np.stack(orig_ious, 0).mean(0)
orig_iouss.extend(orig_ious)
seq_ious.extend(seq_iou)
print(cat, orig_ious, seq_iou)
print(sum(orig_iouss)/len(orig_iouss))
print(sum(seq_ious)/len(seq_ious))
if __name__ == '__main__':
parser = ArgumentParser()
parser.add_argument('--dataset', type=str, choices=['dvs16', 'dvs17m', 'dvs17'], default='dvs17')
parser.add_argument('--pred_dir', type=str, default='')
parser.add_argument('--flow_pred_dir', type=str, default='')
parser.add_argument('--save_path', type=str, default='seq_masks')
args = parser.parse_args()
args.flow_b1_dir = ''
args.flow_f1_dir= ''
args.flow_b2_dir = ''
args.flow_f2_dir= ''
if args.dataset == 'dvs17' or args.dataset == 'dvs17m'
args.cats = ['bike-packing', 'blackswan', 'bmx-trees', 'breakdance', 'camel', 'car-roundabout', 'car-shadow',
'cows', 'dance-twirl', 'dog', 'dogs-jump', 'drift-chicane', 'drift-straight', 'goat', 'gold-fish',
'horsejump-high', 'india', 'judo', 'kite-surf', 'lab-coat', 'libby', 'loading', 'mbike-trick',
'motocross-jump', 'paragliding-launch', 'parkour', 'pigs', 'scooter-black', 'shooting', 'soapbox']
else:
args.cats = ['dog', 'cows', 'goat', 'camel', 'libby', 'parkour', 'soapbox', 'blackswan', 'bmx-trees',
'kite-surf', 'car-shadow', 'breakdance', 'dance-twirl', 'scooter-black', 'drift-chicane',
'motocross-jump', 'horsejump-high', 'drift-straight', 'car-roundabout', 'paragliding-launch']
if args.dataset == 'dvs17':
args.obj_nums = {'bike-packing': 2, 'blackswan': 1, 'bmx-trees': 2, 'breakdance': 1, 'camel': 1, 'car-roundabout': 1, 'car-shadow':1 ,
'cows': 1, 'dance-twirl': 1, 'dog': 1, 'dogs-jump': 3, 'drift-chicane': 1, 'drift-straight': 1, 'goat': 1, 'gold-fish': 5,
'horsejump-high': 2, 'india': 8, 'judo': 2, 'kite-surf': 3, 'lab-coat': 5, 'libby': 1, 'loading': 3, 'mbike-trick': 2,
'motocross-jump': 2, 'paragliding-launch': 3, 'parkour': 1, 'pigs': 3, 'scooter-black': 2, 'shooting': 2, 'soapbox': 3}
args.gt_dir = ''
else:
args.obj_nums = {'bike-packing': 2, 'blackswan': 1, 'bmx-trees': 1, 'breakdance': 1, 'camel': 1, 'car-roundabout': 1, 'car-shadow':1 ,
'cows': 1, 'dance-twirl': 1, 'dog': 1, 'dogs-jump': 3, 'drift-chicane': 1, 'drift-straight': 1, 'goat': 1, 'gold-fish': 5,
'horsejump-high': 1, 'india': 3, 'judo': 2, 'kite-surf': 1, 'lab-coat': 1, 'libby': 1, 'loading': 3, 'mbike-trick': 1,
'motocross-jump': 1, 'paragliding-launch': 1, 'parkour': 1, 'pigs': 3, 'scooter-black': 1, 'shooting': 1, 'soapbox': 1}
args.gt_dir = ''
main(args)