Training on Cityscapes

[Jinming-Su]

I resize all image 1024*512,but when I start training I get this error:

I0423 14:01:20.146436  3294 solver.cpp:251] Learning Rate Policy: step
*** Error in `../../caffe-segnet/build/tools/caffe': free(): invalid next size (fast): 0x0000000003375a60 ***
*** Aborted at 1492927358 (unix time) try "date -d @1492927358" if you are using GNU date ***
PC: @     0x7f869b32cc37 (unknown)
*** SIGABRT (@0x3e800000cde) received by PID 3294 (TID 0x7f869c6ae9c0) from PID 3294; stack trace: ***
    @     0x7f869b32ccb0 (unknown)
    @     0x7f869b32cc37 (unknown)
    @     0x7f869b330028 (unknown)
    @     0x7f869b3692a4 (unknown)
    @     0x7f869b37555e (unknown)
    @     0x7f869c19f4ac caffe::AccuracyLayer<>::Forward_cpu()
    @     0x7f869c24323a caffe::Net<>::ForwardFromTo()
    @     0x7f869c2434d7 caffe::Net<>::ForwardPrefilled()
    @     0x7f869c25adc7 caffe::Solver<>::Step()
    @     0x7f869c25b6ff caffe::Solver<>::Solve()
    @           0x4071c8 train()
    @           0x405771 main
    @     0x7f869b317f45 (unknown)
    @           0x405d1d (unknown)
    @                0x0 (unknown)
Aborted (core dumped)

The why?

[Jinming-Su]

solved by change the output_num

[muralabmahmuds]

Dear @su526664687

I am also currently studying on SegNet and I want to train it using Cityscapes. Anyway, what about your result after resizing the image dimension into 512x1024? Can you share your prototxt (train, inference, and solver) which you used? I am confused which parts should be modified if I use Cityscapes dataset.

Thank you.