function rcnn_model = rcnn_create_model(cnn_definition_file, cnn_binary_file, cache_name)

% AUTORIGHTS

% ---------------------------------------------------------

% Copyright (c) 2014, Ross Girshick

%

% This file is part of the R-CNN code and is available

% under the terms of the Simplified BSD License provided in

% LICENSE. Please retain this notice and LICENSE if you use

% this file (or any portion of it) in your project.

% ---------------------------------------------------------

if ~exist('cache_name', 'var') || isempty(cache_name)

cache_name = 'none';

end

% model =

% cnn: [1x1 struct]

% binary_file: 'path/to/cnn/model/binary'

% definition_file: 'path/to/cnn/model/definition'

% batch_size: 256

% image_mean: [227x227x3 single]

% init_key: -1

% detectors.W: [N x <numclasses> single] % matrix of SVM weights

% detectors.B: [1 x <numclasses> single] % (row) vector of SVM biases

% detectors.crop_mode: 'warp' or 'square'

% detectors.crop_padding: 16

% detectors.nms_thresholds: [1x20 single]

% training_opts: [1x1 struct]

% bias_mult: 10

% fine_tuned: 1

% layer: 'fc7'

% pos_loss_weight: 2

% svm_C: 1.0000e-03

% trainset: 'trainval'

% use_flipped: 0

% year: '2007'

% feat_norm_mean: 20.1401

% classes: {cell array of class names}

% class_to_index: map from class name to column index in W

% init empty convnet

assert(exist(cnn_binary_file, 'file') ~= 0);

assert(exist(cnn_definition_file, 'file') ~= 0);

cnn.binary_file = cnn_binary_file;

cnn.definition_file = cnn_definition_file;

cnn.batch_size = 256;

cnn.init_key = -1;

cnn.input_size = 227;

% load the ilsvrc image mean

data_mean_file = './external/caffe/matlab/caffe/ilsvrc_2012_mean.mat';

assert(exist(data_mean_file, 'file') ~= 0);

ld = load(data_mean_file);

image_mean = ld.image_mean; clear ld;

off = floor((size(image_mean,1) - cnn.input_size)/2)+1;

image_mean = image_mean(off:off+cnn.input_size-1, off:off+cnn.input_size-1, :);

cnn.image_mean = image_mean;

% init empty detectors

detectors.W = [];

detectors.B = [];

detectors.crop_mode = 'warp';

detectors.crop_padding = 16;

detectors.nms_thresholds = [];

% rcnn model wraps the convnet and detectors

rcnn_model.cnn = cnn;

rcnn_model.cache_name = cache_name;

rcnn_model.detectors = detectors;