Shih-En Wei

The caffe was forked from earlier versions, so it may be not compatible with opencv 3. The long term solution would be merging our caffe with more recent caffe, but...

I was using cuda 7.5 and cudnn 3.

Hi, could you check the batch size? I was using 4 or 8 on 1 GPU.

No I using batch size at least 4 on each GPU. You are right, that code has problems with batch size 1. Will update.

We've tried adding BN layers after each CONV layers, but didn't get better convergence speed and accuracy. But there's a big speedup with VGG-pretrained model vs. training from scratch. Given...

Fine-tuning is pretty much the same as training from scratch: you prepare a LMDB. With a new dataset, you need to look into `training/genJSON.m` and write your own code on...

The lmdbs will be around 110 GB ~ 130 GB if MPI dataset is included. Keep this issue open if you still find errors with unmodified genLMDB.py.