今天我在跑crpn的时候，遇见了如下错误：

I0416 22:15:28.233808 17888 net.cpp:744] Ignoring source layer loss_rpn_tl
I0416 22:15:28.233809 17888 net.cpp:744] Ignoring source layer loss_rpn_tr
I0416 22:15:28.233811 17888 net.cpp:744] Ignoring source layer loss_rpn_br
I0416 22:15:28.233814 17888 net.cpp:744] Ignoring source layer loss_rpn_bl
I0416 22:15:28.233815 17888 net.cpp:744] Ignoring source layer roi-data
I0416 22:15:28.308246 17888 net.cpp:744] Ignoring source layer loss_cls
I0416 22:15:28.308260 17888 net.cpp:744] Ignoring source layer loss_bbox
im_detect: 1/1000 0.309s 0.001s
F0416 22:15:28.995488 17888 syncedmem.cpp:71] Check failed: error == cudaSuccess (2 vs. 0)  out of memory
*** Check failure stack trace: ***
./experiments/scripts/test.sh: line 48: 17888 Aborted                 (core dumped) ./tools/test_net.py --gpu 0 --def models/${NET}/test.pt --net ${MODEL} --imdb ${TEST_IMDB} --cfg models/${NET}/config.yml ${EXTRA_ARGS}

c 查资料发现，原来是编译faster-rcnn的时候，忘记用cudnn了，于是到

cd caffe-fast-rcnn

修改把caffe配置文件Makefile.config的cudnn那项的注释解除：

# cuDNN acceleration switch (uncomment to build with cuDNN).
USE_CUDNN := 1
make clean
make -j8 && make pycaffe

 参考文献

[1].Check failed: error == cudaSuccess (2 vs. 0) out of memory. 

​​https://github.com/CMU-Perceptual-Computing-Lab/caffe_rtpose/issues/44​​

[2]. Corner-based Region Proposal Network. https://github.com/xhzdeng/crpn