how to solve the lib reimplement for minibatch size >1 error?

[ygren]

hello，
This project implements end to end training with minibatch = 1，it takes up about 4G~5G GPU resources .but my single GPU resources is 24G, and my dataset is about 250k images. It's very wasteful.I try to reimplement the lib with this repo [https://github.com/ice-pice/py-faster-rcnn] , but this implementation is for Faster RCNN python version. I have checked your commit history,the error occurred when you added TRAINAGNOSTIC operation.
Traceback (most recent call last): File "./tools/train_net.py", line 112, in <module> max_iters=args.max_iters) File "/home/ryg/FV/py-R-FCN/tools/../lib/fast_rcnn/train.py", line 157, in train_net pretrained_model=pretrained_model) File "/home/ryg/FV/py-R-FCN/tools/../lib/fast_rcnn/train.py", line 43, in __init__ self.solver = caffe.SGDSolver(solver_prototxt) File "/home/ryg/FV/py-R-FCN/tools/../lib/roi_data_layer/layer.py", line 130, in setup top[idx].reshape(1) IndexError: Index out of range
why the blob index out of range.
Thanks and looking forward to your reply.