Is it possible to extract features using 11GB GPU?

ChenRocks / UNITER

Research code for ECCV 2020 paper "UNITER: UNiversal Image-TExt Representation Learning"

https://arxiv.org/abs/1909.11740

777 stars 109 forks source link

Is it possible to extract features using 11GB GPU? #78

Open VisualJoyce opened 3 years ago

VisualJoyce commented 3 years ago

Now, the process reports F0625 08:25:59.166216 119 syncedmem.cpp:71] Check failed: error == cudaSuccess (2 vs. 0) out of memory on my 11GB 2080ti. But that's all we have.

VisualJoyce commented 3 years ago

When I load the model into GPU, it costs only 1.8G, but the feature extraction requires way larger than that.