function [scores, list_im]= feat_batch (use_gpu, net_model, net_weights, list_im, dim)

caffe.reset_all();

if nargin < 1

% By default use CPU

use_gpu = 0;

if ischar(list_im)

%Assume it is a file contaning the list of images

filename = list_im;

list_im = read_cell(filename);

if exist('use_gpu', 'var') && use_gpu

caffe.set_mode_gpu();

gpu_id = 0; % we will use the first gpu in this demo

caffe.set_device(gpu_id);

caffe.set_mode_cpu();

phase = 'test'; % run with phase test (so that dropout isn't applied)

if ~exist(net_weights, 'file')

error('%s does not exist.', net_weights);

if ~exist(net_model, 'file')

error('%s does not exist.', net_model);

net = caffe.Net(net_model, net_weights, phase);

d = load('./matlab/+caffe/imagenet/ilsvrc_2012_mean.mat');

mean_data = d.mean_data;

batch_size = 10;

num_images = length(list_im);

scores = zeros(dim,num_images,'single');

num_batches = ceil(length(list_im)/batch_size);

initic=tic;

for bb = 1 : num_batches

batchtic = tic;

range = 1+batch_size*(bb-1):min(num_images,batch_size * bb);

tic

input_data = {prepare_batch(list_im(range),mean_data,batch_size)};

toc, tic

fprintf('Batch %d out of %d %.2f%% Complete ETA %.2f seconds\n',...

bb,num_batches,bb/num_batches*100,toc(initic)/bb*(num_batches-bb));

%output_data = caffe('forward', {input_data});

output_data = net.forward(input_data);

toc

output_data = squeeze(output_data{1});

scores(:,range) = output_data(:,mod(range-1,batch_size)+1);

toc(batchtic)

toc(initic);

caffe.reset_all();