-
Notifications
You must be signed in to change notification settings - Fork 2
/
cityscapes_pred.py
184 lines (168 loc) · 12.4 KB
/
cityscapes_pred.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
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
import os
import cv2
import argparse
import numpy as np
import torch
from torch import nn
from torchvision import transforms
from torchvision.models import mobilenet_v2
from cityscapes2 import Cityscapes
parser = argparse.ArgumentParser(description='PyTorch Training Script')
parser.add_argument('--no-cuda', action='store_true', default=False,
help='disables CUDA training')
parser.add_argument('--seed', type=int, default=1, metavar='S',
help='random seed (default: 1)')
torch.cuda.empty_cache()
args = parser.parse_args()
args.cuda = not args.no_cuda and torch.cuda.is_available()
torch.manual_seed(args.seed)
if args.cuda:
torch.cuda.manual_seed(args.seed)
cell_list = [[[512, 320], [576, 352]], [[576, 320], [640, 352]], [[640, 320], [704, 352]], [[704, 320], [768, 352]],
[[768, 320], [832, 352]], [[832, 320], [896, 352]], [[896, 320], [960, 352]], [[960, 320], [1024, 352]],
[[1024, 320], [1088, 352]], [[1088, 320], [1152, 352]], [[1152, 320], [1216, 352]],
[[1216, 320], [1280, 352]], [[1280, 320], [1344, 352]], [[1344, 320], [1408, 352]],
[[1408, 320], [1472, 352]], [[1472, 320], [1536, 352]], [[512, 352], [576, 384]], [[576, 352], [640, 384]],
[[640, 352], [704, 384]], [[704, 352], [768, 384]], [[768, 352], [832, 384]], [[832, 352], [896, 384]],
[[896, 352], [960, 384]], [[960, 352], [1024, 384]], [[1024, 352], [1088, 384]],
[[1088, 352], [1152, 384]], [[1152, 352], [1216, 384]], [[1216, 352], [1280, 384]],
[[1280, 352], [1344, 384]], [[1344, 352], [1408, 384]], [[1408, 352], [1472, 384]],
[[1472, 352], [1536, 384]], [[512, 384], [576, 416]], [[576, 384], [640, 416]], [[640, 384], [704, 416]],
[[704, 384], [768, 416]], [[768, 384], [832, 416]], [[832, 384], [896, 416]], [[896, 384], [960, 416]],
[[960, 384], [1024, 416]], [[1024, 384], [1088, 416]], [[1088, 384], [1152, 416]],
[[1152, 384], [1216, 416]], [[1216, 384], [1280, 416]], [[1280, 384], [1344, 416]],
[[1344, 384], [1408, 416]], [[1408, 384], [1472, 416]], [[1472, 384], [1536, 416]],
[[512, 416], [576, 448]], [[576, 416], [640, 448]], [[640, 416], [704, 448]], [[704, 416], [768, 448]],
[[768, 416], [832, 448]], [[832, 416], [896, 448]], [[896, 416], [960, 448]], [[960, 416], [1024, 448]],
[[1024, 416], [1088, 448]], [[1088, 416], [1152, 448]], [[1152, 416], [1216, 448]],
[[1216, 416], [1280, 448]], [[1280, 416], [1344, 448]], [[1344, 416], [1408, 448]],
[[1408, 416], [1472, 448]], [[1472, 416], [1536, 448]], [[512, 448], [576, 480]], [[576, 448], [640, 480]],
[[640, 448], [704, 480]], [[704, 448], [768, 480]], [[768, 448], [832, 480]], [[832, 448], [896, 480]],
[[896, 448], [960, 480]], [[960, 448], [1024, 480]], [[1024, 448], [1088, 480]],
[[1088, 448], [1152, 480]], [[1152, 448], [1216, 480]], [[1216, 448], [1280, 480]],
[[1280, 448], [1344, 480]], [[1344, 448], [1408, 480]], [[1408, 448], [1472, 480]],
[[1472, 448], [1536, 480]], [[512, 480], [576, 512]], [[576, 480], [640, 512]], [[640, 480], [704, 512]],
[[704, 480], [768, 512]], [[768, 480], [832, 512]], [[832, 480], [896, 512]], [[896, 480], [960, 512]],
[[960, 480], [1024, 512]], [[1024, 480], [1088, 512]], [[1088, 480], [1152, 512]],
[[1152, 480], [1216, 512]], [[1216, 480], [1280, 512]], [[1280, 480], [1344, 512]],
[[1344, 480], [1408, 512]], [[1408, 480], [1472, 512]], [[1472, 480], [1536, 512]],
[[512, 512], [576, 544]], [[576, 512], [640, 544]], [[640, 512], [704, 544]], [[704, 512], [768, 544]],
[[768, 512], [832, 544]], [[832, 512], [896, 544]], [[896, 512], [960, 544]], [[960, 512], [1024, 544]],
[[1024, 512], [1088, 544]], [[1088, 512], [1152, 544]], [[1152, 512], [1216, 544]],
[[1216, 512], [1280, 544]], [[1280, 512], [1344, 544]], [[1344, 512], [1408, 544]],
[[1408, 512], [1472, 544]], [[1472, 512], [1536, 544]], [[512, 544], [576, 576]], [[576, 544], [640, 576]],
[[640, 544], [704, 576]], [[704, 544], [768, 576]], [[768, 544], [832, 576]], [[832, 544], [896, 576]],
[[896, 544], [960, 576]], [[960, 544], [1024, 576]], [[1024, 544], [1088, 576]],
[[1088, 544], [1152, 576]], [[1152, 544], [1216, 576]], [[1216, 544], [1280, 576]],
[[1280, 544], [1344, 576]], [[1344, 544], [1408, 576]], [[1408, 544], [1472, 576]],
[[1472, 544], [1536, 576]], [[512, 576], [576, 608]], [[576, 576], [640, 608]], [[640, 576], [704, 608]],
[[704, 576], [768, 608]], [[768, 576], [832, 608]], [[832, 576], [896, 608]], [[896, 576], [960, 608]],
[[960, 576], [1024, 608]], [[1024, 576], [1088, 608]], [[1088, 576], [1152, 608]],
[[1152, 576], [1216, 608]], [[1216, 576], [1280, 608]], [[1280, 576], [1344, 608]],
[[1344, 576], [1408, 608]], [[1408, 576], [1472, 608]], [[1472, 576], [1536, 608]],
[[512, 608], [576, 640]], [[576, 608], [640, 640]], [[640, 608], [704, 640]], [[704, 608], [768, 640]],
[[768, 608], [832, 640]], [[832, 608], [896, 640]], [[896, 608], [960, 640]], [[960, 608], [1024, 640]],
[[1024, 608], [1088, 640]], [[1088, 608], [1152, 640]], [[1152, 608], [1216, 640]],
[[1216, 608], [1280, 640]], [[1280, 608], [1344, 640]], [[1344, 608], [1408, 640]],
[[1408, 608], [1472, 640]], [[1472, 608], [1536, 640]], [[0, 640], [128, 704]], [[128, 640], [256, 704]],
[[256, 640], [384, 704]], [[384, 640], [512, 704]], [[512, 640], [640, 704]], [[640, 640], [768, 704]],
[[768, 640], [896, 704]], [[896, 640], [1024, 704]], [[1024, 640], [1152, 704]],
[[1152, 640], [1280, 704]], [[1280, 640], [1408, 704]], [[1408, 640], [1536, 704]],
[[1536, 640], [1664, 704]], [[1664, 640], [1792, 704]], [[1792, 640], [1920, 704]],
[[1920, 640], [2048, 704]], [[0, 704], [128, 768]], [[128, 704], [256, 768]], [[256, 704], [384, 768]],
[[384, 704], [512, 768]], [[512, 704], [640, 768]], [[640, 704], [768, 768]], [[768, 704], [896, 768]],
[[896, 704], [1024, 768]], [[1024, 704], [1152, 768]], [[1152, 704], [1280, 768]],
[[1280, 704], [1408, 768]], [[1408, 704], [1536, 768]], [[1536, 704], [1664, 768]],
[[1664, 704], [1792, 768]], [[1792, 704], [1920, 768]], [[1920, 704], [2048, 768]], [[0, 768], [128, 832]],
[[128, 768], [256, 832]], [[256, 768], [384, 832]], [[384, 768], [512, 832]], [[512, 768], [640, 832]],
[[640, 768], [768, 832]], [[768, 768], [896, 832]], [[896, 768], [1024, 832]], [[1024, 768], [1152, 832]],
[[1152, 768], [1280, 832]], [[1280, 768], [1408, 832]], [[1408, 768], [1536, 832]],
[[1536, 768], [1664, 832]], [[1664, 768], [1792, 832]], [[1792, 768], [1920, 832]],
[[1920, 768], [2048, 832]], [[0, 832], [128, 896]], [[128, 832], [256, 896]], [[256, 832], [384, 896]],
[[384, 832], [512, 896]], [[512, 832], [640, 896]], [[640, 832], [768, 896]], [[768, 832], [896, 896]],
[[896, 832], [1024, 896]], [[1024, 832], [1152, 896]], [[1152, 832], [1280, 896]],
[[1280, 832], [1408, 896]], [[1408, 832], [1536, 896]], [[1536, 832], [1664, 896]],
[[1664, 832], [1792, 896]], [[1792, 832], [1920, 896]], [[1920, 832], [2048, 896]], [[0, 896], [128, 960]],
[[128, 896], [256, 960]], [[256, 896], [384, 960]], [[384, 896], [512, 960]], [[512, 896], [640, 960]],
[[640, 896], [768, 960]], [[768, 896], [896, 960]], [[896, 896], [1024, 960]], [[1024, 896], [1152, 960]],
[[1152, 896], [1280, 960]], [[1280, 896], [1408, 960]], [[1408, 896], [1536, 960]],
[[1536, 896], [1664, 960]], [[1664, 896], [1792, 960]], [[1792, 896], [1920, 960]],
[[1920, 896], [2048, 960]], [[0, 960], [128, 1024]], [[128, 960], [256, 1024]], [[256, 960], [384, 1024]],
[[384, 960], [512, 1024]], [[512, 960], [640, 1024]], [[640, 960], [768, 1024]], [[768, 960], [896, 1024]],
[[896, 960], [1024, 1024]], [[1024, 960], [1152, 1024]], [[1152, 960], [1280, 1024]],
[[1280, 960], [1408, 1024]], [[1408, 960], [1536, 1024]], [[1536, 960], [1664, 1024]],
[[1664, 960], [1792, 1024]], [[1792, 960], [1920, 1024]], [[1920, 960], [2048, 1024]]]
interested_classes = [(11, 12), (13, 14, 15, 16, 17, 18),
(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 25, 21, 20, 19), (0, 23, 22, 24)]
NumCell = 256 # number of cells
NumClass = 3 # number of classes except background class
root = "Datasets/Cityscapes"
save_name = 'Cityscapes_mobilenet' # name of the model
model = mobilenet_v2() # load the model
model.classifier[1] = nn.Linear(1280, NumCell * (NumClass + 1))
model.load_state_dict(torch.load('Models/' + save_name + '.pth')) # load the model
class_names = ["People", "Vehicle", "Other", "Road", "Background"]
color_box = [(31, 112, 255), (151, 157, 255), (56, 56, 255)]
val_dataset = Cityscapes(root, cell_list=cell_list, interested_classes=interested_classes, split='val',
target_type='semantic')
def pred_plot(frame, original, output):
frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
output = output.detach().numpy()
pred = np.where(output > 0.5, 1, 0).tolist()
cell = 0
normal = np.asarray([0] * NumClass + [1])
for rect in cell_list:
x1, y1 = rect[0]
x2, y2 = rect[1]
each = pred[cell:cell + NumClass + 1]
eachScore = output[cell:cell + NumClass + 1]
# each = orig[cell:cell + NumClass + 1]
if not (each == normal).all():
index = [i for i, x in enumerate(each) if x == 1]
if len(index) == 0:
index.append(eachScore.argmax())
if eachScore.argmax() == NumClass:
continue
center_x, center_y = (x1 + x2) // 2, (y1 + y2) // 2 # 计算矩形中心点
poly_area = (x2 - x1) * (y2 - y1) # 计算矩形面积
default_text_scale = 0.4 # 这是默认的字体大小，可以根据你的需要进行调整
texts = [] # 用于存储所有的文本和对应的大小
max_text_len = len(index) # 计算最长的文本长度
if max_text_len > 1:
text_scale = default_text_scale * min(1, np.sqrt(poly_area) / max_text_len)
else:
text_scale = min(max(poly_area / 10000, 0.3), 0.6)
for i in index:
text_size, _ = cv2.getTextSize(class_names[i], cv2.FONT_HERSHEY_SIMPLEX, text_scale, 2)
texts.append((class_names[i], text_size, text_scale))
text_origin = [center_x, center_y - sum(text[1][1] for text in texts) // 2]
line_spacing = 0.7 # 行间距，可以根据需要调整
color = color_box[index[0]]
for text, text_size, text_scale in texts:
text_origin[0] = center_x - text_size[0] // 2 # 每行的x坐标需要重新计算以保证居中
text_origin[1] += int(text_size[1] * line_spacing) # y坐标加上当前行文本的高度的一部分
# cv2.putText(frame, text, tuple(text_origin), cv2.FONT_HERSHEY_SIMPLEX, text_scale, (255, 255, 255), 1)
text_origin[1] += int(text_size[1] * line_spacing) # y坐标再加上当前行文本的高度的一部分，为下一行文本做准备
cv2.rectangle(frame, tuple(rect[0]), tuple(rect[1]), color=color, thickness=3)
cell += NumClass + 1
return frame
preprocess = transforms.Compose([transforms.ToTensor()])
def test():
model.eval()
with torch.no_grad():
for batch_idx, (image, target, filename) in enumerate(val_dataset):
resize_image = cv2.resize(image, (224, 224))
input_tensor = preprocess(resize_image)
input_batch = input_tensor.unsqueeze(0)
output = model(input_batch)
output = torch.sigmoid(output)
target = torch.tensor(target, dtype=torch.float32)
frame = pred_plot(image, torch.Tensor.cpu(target), output[0])
cv2.imwrite(os.path.join(path, filename + ".jpg"), frame)
current_path = os.getcwd()
path = os.path.join(current_path, save_name)
if not os.path.exists(path):
os.makedirs(path)
test()