如何转换为多分类模型？

Question

我正在尝试使用云样本github实现多类别分类，这是一个分类模型，我必须更改代码。我发现了一些建议，可以将最终层和损失从softmax更改为Sigmoid。此外，我还必须更改标签到一种热编码。有人可以帮助将标签更改为一种热编码吗？

Answer 1

Please refer to the following article:

https://towardsdatascience.com/multi-label-image-classification-with-inception-net-cbb2ee538e30

The technique used in that article looks something like this:

ground_truth = np.zeros(class_count, dtype=np.float32)
ground_truth[label_index] = 1.0

This may not scale well for a very large number of classes (10s of thousands). To scale to higher numbers of classes, you need the equivalent of tf.nn.sparse_softmax_cross_entropy_with_logits for sigmoid, which doesn't seem to exist.