arm linux系统使用paddlelite部署yolov5(python demo)_yolov5在arm

arm linux系统使用paddlelite部署yolov5(python demo)

1.编译环境

arm linux系统利用源码编译paddlelite
详见link

# 下载 Paddle Lite 源码并切换到发布分支，如 release/v2.10
git clone https://github.com/PaddlePaddle/Paddle-Lite.git
cd Paddle-Lite && git checkout release/v2.10

# (可选) 删除 third-party 目录，编译脚本会自动从国内 CDN 下载第三方库文件
# rm -rf third-party

# 执行编译脚本
./lite/tools/build_linux.sh
1
2
3
4
5
6
7
8
9

编译成功后，会在Paddle-Lite/build.lite.linux.armv8.gcc/inference_lite_lib.armlinux.armv8/
python/install/dist 目录下生成对应的.whl轮子，直接用pip安装即可。

模型转换

安装x2paddle

pip install x2paddle
1

以onnx模型为例，更多查看link

from x2paddle.convert import onnx2paddle

model_path = "/home/server/桌面/paddle_demo/arm/yolov5s_sim.onnx"
save_dir = "./"

onnx2paddle(model_path, save_dir,
            convert_to_lite=True,
            lite_valid_places="arm",
            lite_model_type="naive_buffer")

# model_path(str) 为 ONNX 模型路径
# save_dir(str) 为转换后模型保存路径
# convert_to_lite(bool) 表示是否使用 opt 工具，默认为 False

# lite_valid_places(str) 指定转换类型，默认为 arm
# lite_valid_places参数目前可支持 arm、 opencl、 x86、 metal、 xpu、 bm、 mlu、
# intel_fpga、 huawei_ascend_npu、imagination_nna、
# rockchip_npu、 mediatek_apu、 huawei_kirin_npu、 amlogic_npu，可以同时指定多个硬件平台
# (以逗号分隔，优先级高的在前)，opt 将会自动选择最佳方式。

# lite_model_type(str) 指定模型转化类型，目前支持两种类型：protobuf 和 naive_buffer，默认为 naive_buffer
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21

生成的.nb文件即为paddlelite的推理模型

利用paddlelite进行yolov5推理

from paddlelite.lite import *
import numpy as np
import cv2
import time

CLASSES=[]




def pynms(dets, thresh): #非极大抑制
    x1 = dets[:, 0]
    y1 = dets[:, 1]
    x2 = dets[:, 2]
    y2 = dets[:, 3]
    areas = (y2 - y1 + 1) * (x2 - x1 + 1)
    scores = dets[:, 4]
    keep = []
    index = scores.argsort()[::-1] #置信度从大到小排序（下标）

    while index.size > 0:
        i = index[0]
        keep.append(i)

        x11 = np.maximum(x1[i], x1[index[1:]])  # 计算相交面积
        y11 = np.maximum(y1[i], y1[index[1:]])
        x22 = np.minimum(x2[i], x2[index[1:]])
        y22 = np.minimum(y2[i], y2[index[1:]])

        w = np.maximum(0, x22 - x11 + 1)  # 当两个框不想交时x22 - x11或y22 - y11 为负数，
                                           # 两框不相交时把相交面积置0
        h = np.maximum(0, y22 - y11 + 1)  #

        overlaps = w * h
        ious = overlaps / (areas[i] + areas[index[1:]] - overlaps)#计算IOU

        idx = np.where(ious <= thresh)[0]  #IOU小于thresh的框保留下来
        index = index[idx + 1]  # 下标以1开始

    return keep


def xywh2xyxy(x):
    # [x, y, w, h] to [x1, y1, x2, y2]
    y = np.copy(x)
    y[:, 0] = x[:, 0] - x[:, 2] / 2
    y[:, 1] = x[:, 1] - x[:, 3] / 2
    y[:, 2] = x[:, 0] + x[:, 2] / 2
    y[:, 3] = x[:, 1] + x[:, 3] / 2
    return y

def filter_box(org_box,conf_thres,iou_thres): #过滤掉无用的框
    org_box=np.squeeze(org_box) #删除为1的维度
    conf = org_box[..., 4] > conf_thres #删除置信度小于conf_thres的BOX
    # print(conf)
    box = org_box[conf == True]
    cls_cinf = box[..., 5:]
    cls = []
    for i in range(len(cls_cinf)):
        cls.append(int(np.argmax(cls_cinf[i])))
    all_cls = list(set(cls))     #删除重复的类别
    output = []
    for i in range(len(all_cls)):
        curr_cls = all_cls[i]
        curr_cls_box = []
        curr_out_box = []
        for j in range(len(cls)):
            if cls[j] == curr_cls:
                box[j][5] = curr_cls #将第6列元素替换为类别下标
                curr_cls_box.append(box[j][:6])   #当前类别的BOX
        curr_cls_box = np.array(curr_cls_box)
        curr_cls_box = xywh2xyxy(curr_cls_box)
        curr_out_box = pynms(curr_cls_box,iou_thres) #经过非极大抑制后输出的BOX下标
        for k in curr_out_box:
            output.append(curr_cls_box[k])  #利用下标取出非极大抑制后的BOX
    output = np.array(output)
    return output
#将图片缩放到模型输入大小
def letterbox_r(im, new_shape=(640, 640), color=(0, 0, 0)):
    # Resize and pad image while meeting stride-multiple constraints
    shape = im.shape[:2]  # current shape [height, width]
    if isinstance(new_shape, int):
        new_shape = (new_shape, new_shape)

    # Scale ratio (new / old)
    r = min(new_shape[0] / shape[0], new_shape[1] / shape[1])

    # Compute padding
    ratio = r, r  # width, height ratios
    new_unpad = int(round(shape[1] * r)), int(round(shape[0] * r))
    dw, dh = new_shape[1] - new_unpad[0], new_shape[0] - new_unpad[1]  # wh padding

    dw /= 2  # divide padding into 2 sides
    dh /= 2

    if shape[::-1] != new_unpad:  # resize
        im = cv2.resize(im, new_unpad, interpolation=cv2.INTER_LINEAR)
    top, bottom = int(round(dh - 0.1)), int(round(dh + 0.1))
    left, right = int(round(dw - 0.1)), int(round(dw + 0.1))
    im = cv2.copyMakeBorder(im, top, bottom, left, right, cv2.BORDER_CONSTANT, value=color)  # add border
    return im, ratio, (dw, dh)
    
#将预测的坐标信息转换回原图尺度
def scale_coords(img1_shape, coords, img0_shape, ratio_pad=None):
    #  img1_shape: 缩放后的图像尺度
    #  coords: 预测的box信息
    #  img0_shape: 缩放前的图像尺度
    #  ratio_pad: 缩放过程中的缩放比例以及pad

    # Rescale coords (xyxy) from img1_shape to img0_shape
    if ratio_pad is None:  # calculate from img0_shape
        gain = max(img1_shape) / max(img0_shape)  # gain  = old / new
        pad = (img1_shape[1] - img0_shape[1] * gain) / 2, (img1_shape[0] - img0_shape[0] * gain) / 2  # wh padding
    else:
        gain = ratio_pad[0][0]
        pad = ratio_pad[1]

    coords[:, [0, 2]] -= pad[0]  # x padding
    coords[:, [1, 3]] -= pad[1]  # y padding
    coords[:, :4] /= gain
    clip_coords(coords, img0_shape)
    return coords

def clip_coords(boxes, img_shape):
    # # Clip bounding xyxy bounding boxes to image shape (height, width)
    # if isinstance(boxes, torch.Tensor):  # faster individually
    #     boxes[:, 0].clamp_(0, shape[1])  # x1
    #     boxes[:, 1].clamp_(0, shape[0])  # y1
    #     boxes[:, 2].clamp_(0, shape[1])  # x2
    #     boxes[:, 3].clamp_(0, shape[0])  # y2
    # else:  # np.array (faster grouped)
    boxes[:, [0, 2]] = boxes[:, [0, 2]].clip(0, img_shape[1])  # x1, x2
    boxes[:, [1, 3]] = boxes[:, [1, 3]].clip(0, img_shape[0])  # y1, y2


def draw(image,box_data):  #画图
    boxes=box_data[...,:4].astype(np.int32) #取整方便画框
    scores=box_data[...,4]
    classes=box_data[...,5].astype(np.int32) #下标取整

    for box, score, cl in zip(boxes, scores, classes):
        top, left, right, bottom = box
        print('class: {}, score: {}'.format(CLASSES[cl], score))
        print('box coordinate left,top,right,down: [{}, {}, {}, {}]'.format(top, left, right, bottom))

        cv2.rectangle(image, (top, left), (right, bottom), (255, 0, 0), 2)
        cv2.putText(image, '{0} {1:.2f}'.format(CLASSES[cl], score),
                    (top, left ),
                    cv2.FONT_HERSHEY_SIMPLEX,
                    0.6, (0, 0, 255), 2)



model_dir='opt.nb'
img_path='./1.jpg'
or_img=cv2.imread(img_path)
se_img=letterbox_r(or_img,(640,640))[0]
img=se_img[:,:,::-1].transpose(2,0,1)  #BGR2RGB和HWC2CHW
img=img.astype(dtype=np.float32)
img/=255.0
img = np.expand_dims(img, axis=0)
# print(img.shape)
# 1. Set config information
config = MobileConfig()
# 2. Set the path to the model generated by opt tools
config.set_model_from_file(model_dir)
# 3. Create predictor by config
predictor = create_paddle_predictor(config)
# print(predictor)

input_tensor = predictor.get_input(0)
input_tensor.from_numpy(img.astype("float32"))
s=time.time()
predictor.run()
output_tensor = predictor.get_output(0)
e=time.time()
print("infer_time:{}".format(e-s))
output_data = output_tensor.numpy()
# print(output_data.shape)
outbox=filter_box(output_data,0.5,0.5)
outbox[..., :4] = scale_coords(se_img.shape, outbox[..., :4], or_img.shape)
draw(or_img,outbox)
cv2.imwrite('./det2.jpg',or_img)
# cv2.imshow('res',or_img)
# cv2.waitKey(0)
# cv2.destroyallwindows()
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
185
186