Can 'csv' format data be used in Caffe Training Proecess?

[刘治来]

Everyone: 

 I have a prob when I want to train caffe in a Facial Emotion Recognition database.The database listed as follow:

The task is to categorize each face based on the emotion shown in the facial expression(1st Col)  in to one of seven categories (0=Angry, 1=Disgust, 2=Fear, 3=Happy, 4=Sad, 5=Surprise, 6=Neutral).

The face image can be descibed in the 2nd col as a vec ,a space-separated pixel values in row major order.

My prob is :How can I convert this format data to lmdb file ?   Matlab or Python code will be appreciated.......THX A LOT!!!

[Jason lee]

you can convert the csv to hd5 by following code:

import os

import numpy as np

from pandas.io.parsers import read_csv

from sklearn.utils import shuffle

import h5py

TRAIN_CSV = 'path/train.csv'

def csv_to_hd5():

    dataframe = read_csv(os.path.expanduser(TRAIN_CSV))

    dataframe['Image'] = dataframe['Image'].apply(lambda  img:np.fromstring(img,sep='   '))

    dataframe = dataframe.dropna()

    data = np.vstack(dataframe['Image'].values)/255

    label = dataframe[dataframe.columns[:-1]].values

    label = (label-48)/48

    data, label = shuffle(data, label, random_state = 0)

    return data, label

if __name__ == '__main__':

    data,label = csv_to_hd5()

    data = data.reshape(-1,1,96,96)

    data_train = data[:-100,:,:,:]

    data_val = data[-100:,:,:,:]

    label = label.reshape(-1,1,1,30)

    label_train =label[:-100,:,:,:]

    label_val = label[-100:,:,:,:]

    fhandle = h5py.File('train.hd5','w')

    fhandle.create_dataset('data',data = data_val,compression='gzip',compression_opts =4)

    fhandle.create_dataset('label', data=label_val, compression='gzip', compression_opts=4)

    fhandle.close()

2017년 7월 6일 목요일 오후 8시 23분 29초 UTC+8, 刘治来 님의 말: