search

zhaoweicai / mscnn

Caffe implementation of our multi-scale object detection framework
404 stars 211 forks source link

out of memory #68

Closed ailh0928 closed 7 years ago

ailh0928 commented 7 years ago

when I trained the MSCNN(mscnn-7s-576-2x), I got a error: out of memory, so I changed the batch size, and I trained the MSCNN model successfully. however, when I test the model I trained (run_mscnn_detection.m), I got the same error: out of memory. Is this related to GPU memory size? I am using ubuntu 14.04 with GPU titan x (12GB memory). All things I did as explained in readme.md. the error is as follow: I0417 23:41:11.172020 9173 net.cpp:816] Ignoring source layer loss_cls I0417 23:41:11.172026 9173 net.cpp:816] Ignoring source layer cls_accuracy I0417 23:41:11.172032 9173 net.cpp:816] Ignoring source layer loss_bbox I0417 23:41:11.172039 9173 net.cpp:816] Ignoring source layer bbox_iou F0417 23:42:35.005249 9173 syncedmem.cpp:56] Check failed: error == cudaSuccess (2 vs. 0) out of memory Check failure stack trace: Killed

zhaoweicai commented 7 years ago

Please have a look at https://github.com/zhaoweicai/mscnn/issues/7 first. Titan X is sufficient to train and test any provided models without changes. When use run_mscnn_detection.m, did you change the image size to 576x1920?

ailh0928 commented 7 years ago

@zhaoweicai thanks a lot. i have installed cudnn, but not use it for making caffe. thank you very much!