wider_face/ssd_pascal.py 运行时候报错，out of memory 但是我GPU内存足够，为什么会报这种错

来源：4-7 Caffe-SSD人脸检测模型训练实操讲解

swag丶

2018-12-05

Memory required for data: 1985822724
…
…
…

I1205 15:51:37.168557 8514 solver.cpp:294] Solving VGG_wider_face_SSD_300x300_train
I1205 15:51:37.168561 8514 solver.cpp:295] Learning Rate Policy: multistep
I1205 15:51:37.171767 8514 blocking_queue.cpp:50] Data layer prefetch queue empty
F1205 15:51:45.878247 8514 syncedmem.cpp:56] Check failed: error == cudaSuccess (2 vs. 0) out of memory
*** Check failure stack trace: ***
@ 0x7f09843245cd google::LogMessage::Fail()
@ 0x7f0984326433 google::LogMessage::SendToLog()
@ 0x7f098432415b google::LogMessage::Flush()
@ 0x7f0984326e1e google::LogMessageFatal::~LogMessageFatal()
@ 0x7f0984b1f7a0 caffe::SyncedMemory::to_gpu()
@ 0x7f0984b1e7c9 caffe::SyncedMemory::mutable_gpu_data()
@ 0x7f0984b3d873 caffe::Blob<>::mutable_gpu_diff()
@ 0x7f0984b9e80b caffe::ConvolutionLayer<>::Backward_gpu()
@ 0x7f098492437b caffe::Net<>::BackwardFromTo()
@ 0x7f09849243cf caffe::Net<>::Backward()
@ 0x7f098491211c caffe::Solver<>::Step()
@ 0x7f0984912b5e caffe::Solver<>::Solve()
@ 0x40d2e8 train()
@ 0x408d14 main
@ 0x7f0982a5e830 __libc_start_main
@ 0x409629 _start
@ (nil) (unknown)