Evaluation batchsize

[ghost]

Can the batchsize in the evaluate.py file be set to 1? My model is quite heavy and uses full resolution images as input. I'm assuming it is not set to 1 because I'm getting memory errors during submission.

[TimJaspers0801]

The batch size during evaluation is set to 1. Unfortunately, this means your model is too big for the GPU used for evaluation. However, I don't see any failed submission... maybe the file was to large to upload to codalab?

[ghost]

That would be great. It would prevent storage of intermediate activations and gradients (required during training) reducing memory usage by 30-50%.

[TimJaspers0801]

alright, I will let you know when it is updated!

[ghost]

Thanks!

[TimJaspers0801]

Done! Hopefully this helps!

[ghost]

Thanks, it works!