Hello, 

我在修改官方的图像识别教程时（从单类别变成二类别）遇到了问题。烦请大家帮助。

我想把原来的LOSS函数改成cross_entropy_loss或者用sigmod 作为loss 函数，但代码报错说我的维度不对。

以下 是我的label 和预测值的label(batch 10)

Estimate_label

Tensor(shape=[10, 2], dtype=float32, place=Place(gpu:0), stop_gradient=False,
[[-2.36834717, -0.94153249],
[-2.02542281, 0.04710546],
[-1.76077676, 0.05066326],
[-2.27432156, -0.73732662],
[-1.19670498, -1.36968899],
[-1.57935870, -1.01269686],
[-0.64907420, 0.49497414],
[-2.88516164, 0.81037402],
[ 2.38728380, -3.08713198],
[ 1.06694150, -1.77655041]])

Label:
Tensor(shape=[10, 2], dtype=float32, place=Place(gpu:0), stop_gradient=True,
[[0., 1.],
[0., 1.],
[1., 0.],
[0., 1.],
[0., 1.],
[0., 1.],
[0., 1.],
[1., 0.],
[0., 1.],
[1., 0.]])

训练代码

    for epoch in range(EPOCH_NUM):
        for batch_id, data in enumerate(train_data_loader()):
            x_data, y_data = data
            # print(y_data)
            img = paddle.to_tensor(x_data)
            # print(img.shape)
            label = paddle.to_tensor(y_data)
            estimate_label = model(img)
            print(estimate_label)
            print(label)
            cross_entropy_loss = paddle.nn.loss.CrossEntropyLoss(
            )
            loss = cross_entropy_loss(estimate_label, label)
            avg_loss = paddle.mean(loss)
            if batch_id % 20 == 0:
                print("epoch: {}, batch_id: {}, loss is: {:.4f}".format(
                    epoch, batch_id, float(avg_loss.numpy())))
            # 反向传播，更新权重，清除梯度
            avg_loss.backward()
            optimizer.step()
            optimizer.clear_grad()

报错：

/usr/lib/python3/dist-packages/apport/report.py:13: DeprecationWarning: the imp module is deprecated in favour of importlib; see the module's documentation for alternative uses
  import fnmatch, glob, traceback, errno, sys, atexit, locale, imp, stat
Traceback (most recent call last):
  File "cat_dog_classifer.py", line 220, in 
    train(my_alex_net, opt)
  File "cat_dog_classifer.py", line 179, in train
    loss = cross_entropy_loss(estimate_label, label)
  File "/home/tesla/.local/lib/python3.8/site-packages/paddle/fluid/dygraph/layers.py", line 930, in __call__
    return self._dygraph_call_func(*inputs, **kwargs)
  File "/home/tesla/.local/lib/python3.8/site-packages/paddle/fluid/dygraph/layers.py", line 915, in _dygraph_call_func
    outputs = self.forward(*inputs, **kwargs)
  File "/home/tesla/.local/lib/python3.8/site-packages/paddle/nn/layer/loss.py", line 397, in forward
    ret = paddle.nn.functional.cross_entropy(
  File "/home/tesla/.local/lib/python3.8/site-packages/paddle/nn/functional/loss.py", line 1731, in cross_entropy
    _, out = _C_ops.softmax_with_cross_entropy(
ValueError: (InvalidArgument) If Attr(soft_label) == false, the axis dimension of Input(Label) should be 1.
  [Hint: Expected labels_dims[axis] == 1UL, but received labels_dims[axis]:2 != 1UL:1.] (at /paddle/paddle/fluid/operators/softmax_with_cross_entropy_op.cc:193)
  [operator < softmax_with_cross_entropy > error]

我应该如何修改呢?！ 谢谢！