ValueError: cannot reshape array of size xx into shape (x,x,x)

ValueError: cannot reshape array of size 571428 into shape (3,351,407)

在训练CTPN的时候，数据集处理的 cv2.dnn.blobFromImage 之后的reshape报的这个错。原因是有一张图像它的通道数乘以宽和高等于571428，不等于3 * 351 * 407，因此不能reshape到（3，351，407）。
算了一下 571428 = 4 * 351 * 407 ，说明这个图莫名其妙地是个4通道的图像。然后断点找了一下这张图，发现它竟然是一个png格式的后缀名jpg的假jpg，多了一个通道。因为png是RGBA，jpg是RGB。
解决：进行图像转换，把它转换为RGB的三通道图。在这

    def __getitem__(self, idx, scale=600, maxScale=900):
        img_name = self.img_names[idx]
        img_path = os.path.join(self.datadir, img_name)
        img = Image.open(img_path)
        img, rate = resize_img(img, scale, maxScale=maxScale)
        rescale_fac = 1/rate

        h, w = img.shape[:2]
        img = cv2.dnn.blobFromImage(
            img, scalefactor=1.0, size=(w, h), swapRB=False, crop=False) # <Mat>
        img = img.reshape(3, h, w)
        # = torch.cuda.FloatTensor if torch.cuda.is_available() else torch.FloatTensor
        img = torch.tensor(img)

        # img = Variable(img).type(Tensor)
        xml_path = os.path.join(self.labelsdir, img_name.split('.')[0]+'.xml')
        gtbox = self.generate_gtboxes(xml_path, rescale_fac)

        feature_size = (int(np.ceil(h/16)), int(np.ceil(w/16)))
        [cls, regr] = cal_rpn((h, w), feature_size, 16, gtbox)

        # regr = [ancher_nums, 3]
        regr = np.hstack([cls.reshape(cls.shape[0], 1), regr])
        regr = torch.from_numpy(regr).float()

        # cls = [1, ancher_nums]
        cls = np.expand_dims(cls, axis=0)
        cls = torch.from_numpy(cls).float()

        return img, cls, regr