madkn / histogramloss Goto Github PK

Hi, I am an research intern from Peking University, recently, I find that you have experiement result on the stanford product dataset. I am wondering whether you can offer me the related c++ layer file and prototxt files to repeat your experiments.

Thanks

In model/cuhk03,the prototxt use slice layer to output 3 slice_datas.In your paper,your are want to create 3 stream sub-network to input 3 parts of person.
I was wonder the slice_dim :1 just mean you slice image in image channel but not in image height? so the slice data are not slice 3 parts of person ?

Hi,

Could you also provide in training data (lmdb) generation script for the reID task? At an initial step, I was trying your python layers on the MNIST dataset I got the following error:

File "HistogramLoss/caffe/build/install/python/distribution_loss_layer.py", line 24, in getPlaces
if x[inx_sorted[0]] > grid[i]:
IndexError: index out of bounds

Can you provide some help?

Thanks,
M.

Hi, could you explain me the purpose using the label shuffling layer ? Is similar to the page 7 in ImageNet-2016-Hikvision