u版yolov5输入大小_yolov5 输入分辨率设置多少合适

u版yolov5:
模型的输入设置为（640，640），输入图片上1920*1080：
3x1080x1920

——等比例缩放——
3x384x672
——backbone——
3x128x48x84和3x256x24x42和3x512x12x21
——Detect——-
transpose
3x48x84x6和3x24x42x6和3x12x21x6
后处理
12096x6和3024x6和756x6
求和得到15876x6
nms
得到目标？x6

再例如768x1024——预处理——480x640
所以caffe中图像预处理没有办法保持一致，只能还采取等比例缩放加补灰边操作。
相关代码：

def letterbox(im, new_shape=(640, 640), color=(114, 114, 114), auto=True, scaleFill=False, scaleup=True, stride=32):
    # 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])
    if not scaleup:  # only scale down, do not scale up (for better val mAP)
        r = min(r, 1.0)

    # Compute padding
    ratio = r, r  # width, height ratios   （0.625,0.625）
    new_unpad = int(round(shape[1] * r)), int(round(shape[0] * r))     #(640,480)
    dw, dh = new_shape[1] - new_unpad[0], new_shape[0] - new_unpad[1]  # wh padding    ##dw=0,dh=160
    if auto:  # minimum rectangle
        dw, dh = np.mod(dw, stride), np.mod(dh, stride)  # dw=0, dh=0
    elif scaleFill:  # stretch
        dw, dh = 0.0, 0.0
        new_unpad = (new_shape[1], new_shape[0])
        ratio = new_shape[1] / shape[1], new_shape[0] / shape[0]  # width, height ratios

    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)
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