How to create a dataset from scratch

[S]

Hello everyone, 

how can I create a dataset in torch starting from my images? 

Let's say I have 1k images and I want to label it in order to do a supervised learning (I know 1k are too few, but it's an example) 

How can I do? 

Thank you

[Eugenio Culurciello]

see an example of dataset creation and use here:

https://github.com/e-lab/torch7-demos/tree/master/train-face-detector-elab

basically all you need is a bunch of directories with objects

[S]

Thank you E. 
I was just starting reading the code of the person detector you wrote. 
I guess it's pretty similar. 

[S]

the one you linked is awesome clear and self-explanatory. 

Thanks again, E. 

[Toru Hironaka]

Thanks for your post, I have a question about torch tensor. The first image loading tensor "imageAll". This tensor's dimension is (41267, 3, 32, 32 as num of images, channel, width, height). However, the dimension of training tensor "trainData.data" and testing tensor "testData.data" is (<numbe or train or test>, 1, 32, 32 as num of images, channel, width, height) while cloning data from imageAll. Why "trainData.data" and "testData.data" dropped channel number to 1 from 3? Is this because the following code only use 'Y' channel for get mean of images or normalize of images I think?

 

Thanks,