Hello, we have encountered the following problems in reproducing your code, and we look forward to receiving your answers. When we use some ImageNet data sets, VOC data sets, MNIST data sets and our own data sets, the loss during training is very small, but during testing, we find that actually different categories are divided into the same category, that is, there is a one-to-many situation between the predicted category and the real category. We don't know where the problem lies. We look forward to your answer. Thank you very much!
Hello, we have encountered the following problems in reproducing your code, and we look forward to receiving your answers. When we use some ImageNet data sets, VOC data sets, MNIST data sets and our own data sets, the loss during training is very small, but during testing, we find that actually different categories are divided into the same category, that is, there is a one-to-many situation between the predicted category and the real category. We don't know where the problem lies. We look forward to your answer. Thank you very much!