Closed pkrush closed 9 years ago
EOS Photo
EOS Liveview
One issue: LeNet needs to be trained with the same exact type of images and AlexNet does not. The cropped images where just a little different switching from the EOS photos to EOS Liveview, a different focal length, object distance, and lighting(flash vs LED). Not much of a change to the naked eye in the cropped image. The Liveview images have a different starting resolution and a different algorithm to make them round.
I am not so sure about the above comment. It does not matter regardless.
I have learned a lot more about how to change network sizes.
The system is taking around a second per coin. This is about the slowest my SUKO MKII coin hopper can go. (You can just lower the input voltage on the hopper if it's DC to do this.)
I have been working with 256x256 images. I tried LeNet with 40,000 plus 28x28 images with great success and surprise. So I need to try 64x64.