Use FloatTensor(3, 256, 256):fill(.5) as image in case of reading error

[zhangtemplar]

I find in very rare case, I still get nil from f:read('a'), it could be a corruption of the disk file or a corruption of RAM. In this case, instead of throwing error, I may want to continue my training process by using a mean image.

[ghost]

Thank you for your pull request and welcome to our community. We require contributors to sign our Contributor License Agreement, and we don't seem to have you on file. In order for us to review and merge your code, please sign up at https://code.facebook.com/cla - and if you have received this in error or have any questions, please drop us a line at [email protected]. Thanks!

[colesbury]

I'm a little worried that this may hide problems with the dataset or data loading code if it doesn't trigger an error. Did you figure out what was wrong with your images? My guess is that they weren't actually JPG or PNG images even if they had that extension.

[zhangtemplar]

@colesbury no, i have check the images which trigger the error, which works fine by loading it in torch. Again, I suspect it is likely to be memory error.