File size: 3,769 Bytes
bc3753a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import torch
import torch.nn.functional as F

import os
import sys
import cv2
import random
import datetime
import math
import argparse
import numpy as np

import scipy.io as sio
import zipfile
from .net_s3fd import s3fd
from .bbox import *


def detect(net, img, device):
    img = img - np.array([104, 117, 123])
    img = img.transpose(2, 0, 1)
    img = img.reshape((1,) + img.shape)

    if 'cuda' in device:
        torch.backends.cudnn.benchmark = True

    img = torch.from_numpy(img).float().to(device)
    BB, CC, HH, WW = img.size()
    with torch.no_grad():
        olist = net(img)

    bboxlist = []
    for i in range(len(olist) // 2):
        olist[i * 2] = F.softmax(olist[i * 2], dim=1)
    olist = [oelem.data.cpu() for oelem in olist]
    for i in range(len(olist) // 2):
        ocls, oreg = olist[i * 2], olist[i * 2 + 1]
        FB, FC, FH, FW = ocls.size()  # feature map size
        stride = 2**(i + 2)    # 4,8,16,32,64,128
        anchor = stride * 4
        poss = zip(*np.where(ocls[:, 1, :, :] > 0.05))
        for Iindex, hindex, windex in poss:
            axc, ayc = stride / 2 + windex * stride, stride / 2 + hindex * stride
            score = ocls[0, 1, hindex, windex]
            loc = oreg[0, :, hindex, windex].contiguous().view(1, 4)
            priors = torch.Tensor([[axc / 1.0, ayc / 1.0, stride * 4 / 1.0, stride * 4 / 1.0]])
            variances = [0.1, 0.2]
            box = decode(loc, priors, variances)
            x1, y1, x2, y2 = box[0] * 1.0
            # cv2.rectangle(imgshow,(int(x1),int(y1)),(int(x2),int(y2)),(0,0,255),1)
            bboxlist.append([x1, y1, x2, y2, score])
    bboxlist = np.array(bboxlist)
    if 0 == len(bboxlist):
        bboxlist = np.zeros((1, 5))

    return bboxlist

def batch_detect(net, imgs, device):
    imgs = imgs - np.array([104, 117, 123])
    imgs = imgs.transpose(0, 3, 1, 2)

    if 'cuda' in device:
        torch.backends.cudnn.benchmark = True

    imgs = torch.from_numpy(imgs).float().to(device)
    BB, CC, HH, WW = imgs.size()
    with torch.no_grad():
        olist = net(imgs)

    bboxlist = []
    for i in range(len(olist) // 2):
        olist[i * 2] = F.softmax(olist[i * 2], dim=1)
    olist = [oelem.data.cpu() for oelem in olist]
    for i in range(len(olist) // 2):
        ocls, oreg = olist[i * 2], olist[i * 2 + 1]
        FB, FC, FH, FW = ocls.size()  # feature map size
        stride = 2**(i + 2)    # 4,8,16,32,64,128
        anchor = stride * 4
        poss = zip(*np.where(ocls[:, 1, :, :] > 0.05))
        for Iindex, hindex, windex in poss:
            axc, ayc = stride / 2 + windex * stride, stride / 2 + hindex * stride
            score = ocls[:, 1, hindex, windex]
            loc = oreg[:, :, hindex, windex].contiguous().view(BB, 1, 4)
            priors = torch.Tensor([[axc / 1.0, ayc / 1.0, stride * 4 / 1.0, stride * 4 / 1.0]]).view(1, 1, 4)
            variances = [0.1, 0.2]
            box = batch_decode(loc, priors, variances)
            box = box[:, 0] * 1.0
            # cv2.rectangle(imgshow,(int(x1),int(y1)),(int(x2),int(y2)),(0,0,255),1)
            bboxlist.append(torch.cat([box, score.unsqueeze(1)], 1).cpu().numpy())
    bboxlist = np.array(bboxlist)
    if 0 == len(bboxlist):
        bboxlist = np.zeros((1, BB, 5))

    return bboxlist

def flip_detect(net, img, device):
    img = cv2.flip(img, 1)
    b = detect(net, img, device)

    bboxlist = np.zeros(b.shape)
    bboxlist[:, 0] = img.shape[1] - b[:, 2]
    bboxlist[:, 1] = b[:, 1]
    bboxlist[:, 2] = img.shape[1] - b[:, 0]
    bboxlist[:, 3] = b[:, 3]
    bboxlist[:, 4] = b[:, 4]
    return bboxlist


def pts_to_bb(pts):
    min_x, min_y = np.min(pts, axis=0)
    max_x, max_y = np.max(pts, axis=0)
    return np.array([min_x, min_y, max_x, max_y])