ResourceExhaustedError (see above for traceback): OOM when allocating tensor with shape[4096]
类似问题 https://github.com/CharlesShang/TFFRCNN/issues/68
参考：
http://blog.csdn.net/qing101hua/article/details/77153222
方案：
take a look at the memory usage of your gpu with
nvidia-smi -l 3
if still free memory that can be used
-> edit lib/fast_rcnn/train.py
~line396 : config.gpu_options.per_process_gpu_memory_fraction = 0.40
edit 0.40 to larger propotion (e.g. 0.9)