A neural network layer that enables training of deep neural networks directly from crowdsourced labels (e.g. from Amazon Mechanical Turk) or, more generally, labels from multiple annotators with different biases and levels of expertise.
I tried to train from the original images and annotations downloaded from your homepage but obtain bad results using crowd layer and doctor net. But I achieved similar results using Majority-vote and DS labels. So I wonder if crowd layer still works when training from a dataset whose most samples lack annotations especially when there are so many annotators but most of them did not label all samples.
I have downloaded the dogs-vs-cats dataset at the following URL.
I was wondering if you could tell me what kind of pre-processing you used to create data_train.npy and data_test.npy for these data.