Cnn_keras converting to float 32

[alinakhay]

Cnn_keras converting to float 32

[gregwchase]

@alinakhay Thank you - great to hear people are getting something out of it!

Regarding your question: scaling the images may make you run into OOM errors (not just with my model, but with others).

Something Keras has implemented is the multi_gpu_model parameter; you can place the line directly above model.compile. Assuming you're using more than one GPU, this might solve the OOM issues. I haven't implemented it yet in EyeNet, but it'll be here soon!

model = multi_gpu_model(model, gpus=nb_gpus)
https://keras.io/utils/#multi_gpu_model

[alinakhay]

Thank you for your response @gregwchase , yes I have done so already, using 256 - it worked and then some other issue came up, which I won't bother you with :)