训练报错

[wangshouxu]

您好，我用StanfordCars数据集训练，报如下错误，找了好久也没找到是什么原因，请看一下

[FlyingMoon-GitHub]

请告知我您所使用的torchvision版本。在我所使用的torchvision 0.2.2中，ColorJitter类的对象调用get_params函数（见第43至第44行）的返回值是torchvision.transforms.transforms.Compose类型的，故程序运行时不会报告此错误。 Please tell me the version of torchvision in your environment. In torchvision 0.2.2, which is the version in my environment, the type of the returning value of a ColorJitter object when calling get_params (In line 43~44) is torchvision.transforms.transforms.Compose. Therefore, there's no such error reported while running the code.

[wangshouxu]

我用的torchvision版本是0.10.0

[wangshouxu]

会与版本问题有关么

[wangshouxu]

color_transform的类型是 ![Uploading 20220105_152641.jpg…]()

[wangshouxu]

您好，确实是跟版本有关系的，这是0.10.0的get_params ![Uploading 20220105_155417.jpg…]()

[FlyingMoon-GitHub]

我查看了PyTorch 0.10.0的官方文档与代码，ColorJitter类的对象调用get_params函数后返回的确实是一个元组。根据以上资料，请您将原始代码中第43至第46行替换为如下内容，您遇到的此问题应当可以得到解决。 I've looked up the official documentaion and code of torchvision 0.10.0, the type of the returning value of a ColorJitter object when calling get_params is tuple indeed. According to references, please replace line 43~46 of the original code into these shown below, this problem you've been facing should be solved.

                idx, brightness, contrast, saturation, hue = trans.get_params(trans.brightness, trans.contrast,
                                                                              trans.saturation, trans.hue)
                for img_no, trans_img in enumerate(trans_imgs):
                    for id in idx:
                        if id == 0 and brightness is not None:
                            trans_imgs[img_no] = F.adjust_brightness(trans_img, brightness)
                        elif id == 1 and contrast is not None:
                            trans_imgs[img_no] = F.adjust_contrast(trans_img, contrast)
                        elif id == 2 and saturation is not None:
                            trans_imgs[img_no] = F.adjust_saturation(trans_img, saturation)
                        elif id == 3 and hue is not None:
                            trans_imgs[img_no] = F.adjust_hue(trans_img, hue)

[wangshouxu]

感谢！！已经调通了