trained_net.forward(cuda_image)

[colorfuldarkgray]

I succeded to train a model with my own data. Now I want to test with new data so I loaded a trained net and tried to predict a single image:

**19 img=img.resize((224,224)) 20 img_t = transform(img) ---> 21 prediction=net.forward(imgt.unsqueeze(0).cuda()) 22 predicted_class = np.argmax(prediction.cpu().detach().numpy()) 23 raw_out = prediction.cpu().detach().numpy()[0]

TypeError: forward() missing 3 required positional arguments: 'label', 'Iter', and 'density'**

It doesn't make sense to provide the label for a prediction. Would you please comment on these missing arguments.

I will appreciate any help or advice to move forward.

Best regards!

[colorfuldarkgray]

我制作了一个几乎相同的模型，除了遮罩层，并从训练好的模型（遮罩和所有）中复制了权重，但准确度下降了。 我知道在训练期间前向传递需要图层掩码，但在推理期间是否需要它们？

在 train_model.py 中，第 112 行说：

out = net(image, label, torch.Tensor([epoch + 1]), density)

在评估期间如何提供标签以及验证数据？

问候

[colorfuldarkgray]

带有验证数据的评估不能收到标签，不好，不正确，必须修改。