forked from AI-liu/Complex-YOLO
-
Notifications
You must be signed in to change notification settings - Fork 0
/
eval.py
149 lines (118 loc) · 5.25 KB
/
eval.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
from __future__ import division
import torch
import torch.nn as nn
from torch.autograd import Variable
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
import time
import cv2
from scipy import misc
from utils import *
def drawRect(img, pt1, pt2, pt3, pt4, color, lineWidth):
cv2.line(img, pt1, pt2, color, lineWidth)
cv2.line(img, pt2, pt3, color, lineWidth)
cv2.line(img, pt3, pt4, color, lineWidth)
cv2.line(img, pt1, pt4, color, lineWidth)
def get_region_boxes(x, conf_thresh, num_classes, anchors, num_anchors):
if x.dim() == 3:
x = x.unsqueeze(0)
assert (x.size(1) == (7 + num_classes) * num_anchors)
nA = num_anchors # num_anchors = 5
nB = x.data.size(0)
nC = num_classes # num_classes = 8
nH = x.data.size(2) # nH 16
nW = x.data.size(3) # nW 32
# Tensors for cuda support
FloatTensor = torch.cuda.FloatTensor if x.is_cuda else torch.FloatTensor
LongTensor = torch.cuda.LongTensor if x.is_cuda else torch.LongTensor
ByteTensor = torch.cuda.ByteTensor if x.is_cuda else torch.ByteTensor
prediction = x.view(nB, nA, 7+num_classes, nH, nW).permute(0, 1, 3, 4, 2).contiguous()
# Get outputs
x = torch.sigmoid(prediction[..., 0]) # Center x
y = torch.sigmoid(prediction[..., 1]) # Center y
w = prediction[..., 2] # Width
h = prediction[..., 3] # Height
pred_conf = torch.sigmoid(prediction[..., 6]) # Conf
pred_cls = torch.sigmoid(prediction[..., 7:]) # Cls pred.
# Calculate offsets for each grid
grid_x = torch.arange(nW).repeat(nH, 1).view([1, 1, nH, nW]).type(FloatTensor)
grid_y = torch.arange(nH).repeat(nW, 1).t().view([1, 1, nH, nW]).type(FloatTensor)
scaled_anchors = FloatTensor([(a_w , a_h ) for a_w, a_h in anchors])
anchor_w = scaled_anchors[:, 0:1].view((1, nA, 1, 1))
anchor_h = scaled_anchors[:, 1:2].view((1, nA, 1, 1))
# Add offset and scale with anchors
pred_boxes = FloatTensor(prediction.shape)
pred_boxes[..., 0] = x.data + grid_x
pred_boxes[..., 1] = y.data + grid_y
pred_boxes[..., 2] = torch.exp(w.data) * anchor_w
pred_boxes[..., 3] = torch.exp(h.data) * anchor_h
pred_boxes[..., 6] = pred_conf
pred_boxes[..., 7:(7 + nC) ] = pred_cls
pred_boxes = convert2cpu(pred_boxes.transpose(0, 1).contiguous().view(-1, (7 + nC))) # torch.Size([2560, 15])
all_boxes = []
for i in range(2560):
if pred_boxes[i][6] > conf_thresh:
all_boxes.append(pred_boxes[i])
# print(pred_boxes[i])
return all_boxes
# classes
# class_list = ['Car', 'Van' , 'Truck' , 'Pedestrian' , 'Person_sitting' , 'Cyclist' , 'Tram' ]
bc = {}
bc['minX'] = 0;
bc['maxX'] = 80;
bc['minY'] = -40;
bc['maxY'] = 40
bc['minZ'] = -2;
bc['maxZ'] = 1.25
for file_i in range(6030,6230):
test_i = str(file_i).zfill(6)
lidar_file = '/home/ai/KITTI/training/velodyne/' + test_i + '.bin'
calib_file = '/home/ai/KITTI/training/calib/' + test_i + '.txt'
label_file = '/home/ai/KITTI/training/label_2/' + test_i + '.txt'
# load target data
calib = load_kitti_calib(calib_file)
target = get_target(label_file, calib['Tr_velo2cam'])
# print(target)
# load point cloud data
a = np.fromfile(lidar_file, dtype=np.float32).reshape(-1, 4)
b = removePoints(a, bc)
rgb_map = makeBVFeature(b, bc, 40 / 512)
misc.imsave('eval_bv.png', rgb_map)
# load trained model and forward
input = torch.from_numpy(rgb_map) # (512, 1024, 3)
input = input.reshape(1, 3, 512, 1024)
model = torch.load('ComplexYOLO_epoch100')
model.cuda()
output = model(input.float().cuda()) # torch.Size([1, 75, 16, 32])
# eval result
conf_thresh = 0.7
nms_thresh = 0.4
num_classes = int(8)
num_anchors = int(5)
img = cv2.imread('eval_bv.png')
all_boxes = get_region_boxes(output, conf_thresh, num_classes, anchors, num_anchors)
for i in range(len(all_boxes)):
pred_img_y = int(all_boxes[i][0] * 1024.0 / 32.0) # 32 cell = 1024 pixels
pred_img_x = int(all_boxes[i][1] * 512.0 / 16.0) # 16 cell = 512 pixels
pred_img_width = int(all_boxes[i][2] * 1024.0 / 32.0) # 32 cell = 1024 pixels
pred_img_height = int(all_boxes[i][3] * 512.0 / 16.0) # 16 cell = 512 pixels
rect_top1 = int(pred_img_y - pred_img_width / 2)
rect_top2 = int(pred_img_x - pred_img_height / 2)
rect_bottom1 = int(pred_img_y + pred_img_width / 2)
rect_bottom2 = int(pred_img_x + pred_img_height / 2)
cv2.rectangle(img, (rect_top1, rect_top2), (rect_bottom1, rect_bottom2), (255, 0, 0), 1)
# for j in range(50):
# if target[j][1] == 0:
# break
# img_y = int(target[j][1] * 1024.0) # 32 cell = 1024 pixels
# img_x = int(target[j][2] * 512.0) # 16 cell = 512 pixels
# img_width = int(target[j][3] * 1024.0) # 32 cell = 1024 pixels
# img_height = int(target[j][4] * 512.0) # 16 cell = 512 pixels
#
# rect_top1 = int(img_y - img_width / 2)
# rect_top2 = int(img_x - img_height / 2)
# rect_bottom1 = int(img_y + img_width / 2)
# rect_bottom2 = int(img_x + img_height / 2)
# cv2.rectangle(img, (rect_top1, rect_top2), (rect_bottom1, rect_bottom2), (0, 0, 255), 1)
misc.imsave('eval_bv' + test_i + '.png', img)