Young
Young
Hi, @Superjeka, There is a discuss in the [here](https://github.com/tonylins/pytorch-mobilenet-v2/issues/6).
"Number of labels must match number of predictions" Please check your labels and randomly generate the negative sample. You can use netscope to visualize your net and to check whether...
I think yes. Since this label means that whether the face can be classified as a face, while the value of this label can be considered as the confidence of...
Which dataset did you use? While, having plenty of faces is a good thing. But you need enough time to train it. In caffe, each iteration updates the weights based...
There are two ways to improved the face alignment, 1) please check whether there is any error in the code caused the deviation. (I believe this is most possible problem,...
Can you specify your questions in detailed? What the meaning of the "rectangle.txt"? The bounding box (ground truth) of the CelebA dataset?
Do you generate the training data and labels by yourself? If so, what's the learning rate of your code?
I cannot find the final version of my code to generate training data. I can provide the initial edition in my laptop. Notice that the order of landmarks is not...
Yes, you are right! The proportion is not suitable. You need to keep the negative face : (positive face and part face) around 3 : 1. But the proportion of...
But does this is same with the original paper? The paper indicated that 'when input layer depth is 0 the underlying conv is the identity function.' I'm just curious that...