How to classify multiple images (5000+)

[Mipso]

I've been using the Caffe classify example with C (https://github.com/BVLC/caffe/blob/master/examples/cpp_classification/classification.cpp) and with python (https://github.com/BVLC/caffe/blob/master/python/classify.py) and they work well. However, when I try to pass in several images at once (I modified classification.cpp to https://github.com/flx42/caffe/commit/be0bff1a84c9e16fb8e8514dc559f2de5ab1a416), it works, but I cannot pass in more than 2943 images at once on C, and more than 1000 images on python because of an Out of memory error.

I am using GPU with a Titan X 12 Giga, so I am surprised that the limit is so low.

Is there a way to classify more images at once ??

Thanks!

[Yossi Biton]

There will always be some memory limit - batch of 3k images sound too much, even for Titan X.

I suggest you to split them into smaller batches (of 1000 images for example) and feed the network with each batch separately.

[Mipso]

Thank you! Would changing the batch size have an impact on how many images I can pass in?