/
perform_evaluation_on_sequence.m
133 lines (106 loc) · 5.74 KB
/
perform_evaluation_on_sequence.m
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
%% Perform evaluation
% Function performs evaluation of the given method on the selected sequence
function [output_results_cell, output_detections] = perform_evaluation_on_sequence(paths, seq, eval_params)
if(nargin < 3)
error('Not enough input parameters');
end
%% Initialize output results
% (num_frames x 4) matrix of RMSE_water, TP, FP, FN
output_results_cell = cell(1, 4);
counter = 1;
% In first cell we store extracted water-mask, while in the second cell
% we store detections list
output_detections = cell(1, 2);
%% Read masks of USV parts
% Read mask that filters out the areas of the USV visible in the image
if(seq.id >= 1 && seq.id <= 9)
boat_parts_mask_l = imread(fullfile(paths.USV_parts_masks, 'kope67_L.png'));
boat_parts_mask_r = imread(fullfile(paths.USV_parts_masks, 'kope67_R.png'));
elseif(seq.id >= 10 && seq.id <= 14)
boat_parts_mask_l = imread(fullfile(paths.USV_parts_masks, 'kope71_L.png'));
boat_parts_mask_r = imread(fullfile(paths.USV_parts_masks, 'kope71_R.png'));
elseif(seq.id >= 15 && seq.id <= 18)
boat_parts_mask_l = imread(fullfile(paths.USV_parts_masks, 'kope75_L.png'));
boat_parts_mask_r = imread(fullfile(paths.USV_parts_masks, 'kope75_R.png'));
elseif(seq.id >= 19 && seq.id <= 26)
boat_parts_mask_l = imread(fullfile(paths.USV_parts_masks, 'kope81_L.png'));
boat_parts_mask_r = imread(fullfile(paths.USV_parts_masks, 'kope81_R.png'));
elseif(seq.id >= 27 && seq.id <= 28)
boat_parts_mask_l = imread(fullfile(paths.USV_parts_masks, 'kope82_L.png'));
boat_parts_mask_r = imread(fullfile(paths.USV_parts_masks, 'kope82_R.png'));
end
%% Loop through the images....
% Get the number of total frames in the sequence
total_frames = seq.end_frame - seq.start_frame;
%% Get calibration details...
% Get calibration file
fs = cv.FileStorage(fullfile(paths.dataset_path, 'video_data', seq.name, 'calibration.yaml'));
% Get resolution of raw image
sim = [fs.imageSize{1}, fs.imageSize{2}];
% Get parameters for rectification
S = cv.stereoRectify(fs.M1, fs.D1, fs.M2, fs.D2, sim, fs.R, fs.T, 'ZeroDisparity', true, 'Alpha', 1);
% Fix rectification bug...
[S, map_L1, map_L2, ~, ~] = rectifyimages_fix(S, fs, sim); % Fix narrow view bug after rectification
f = waitbar(0, 'Initializing sequence...');
for frm_num = seq.start_frame + 1 : seq.end_frame
% update progress bar
if(mod(counter, 10) == 0)
waitbar(counter/total_frames, f, sprintf('Processing sequence %02d...', seq.id));
end
% load segmentation mask
msk = imread(fullfile(paths.output, sprintf('%08dL_pred.png', frm_num)));
%msk = imread(fullfile(paths.output, sprintf('%08dL.png', frm_num)));
% Override possible detections that are a couse of usv parts
[msk_size_y, msk_size_x, ~] = size(msk);
boat_parts_mask_resized_l = imresize(boat_parts_mask_l, [msk_size_y, msk_size_x], 'method', 'nearest');
boat_parts_mask_resized_r = imresize(boat_parts_mask_r, [msk_size_y, msk_size_x], 'method', 'nearest');
for rgb_counter = 1 : 3 % loop through color channels
% extract current color channel
tmp = msk(:, :, rgb_counter);
% change with the corresponding value of the water component
tmp(boat_parts_mask_resized_l == 1) = eval_params.labels(3, rgb_counter);
% update the segmentation mask
msk(:, :, rgb_counter) = tmp;
end
% Load ground truth file
gtl = load(fullfile(paths.ground_truth, sprintf('%08dL.mat', frm_num)));
gtl = gtl.annotations;
% Filter sea-edge
gtl.sea_edge = filter_sea_edge(gtl.sea_edge, eval_params.img_size);
sea_edge_line = gtl.sea_edge;
% create inverse sea mask
tmp_inv_sea_mask = poly2mask([1; sea_edge_line(:,1); eval_params.img_size(2)], [1; sea_edge_line(:,2); 1], eval_params.img_size(1), eval_params.img_size(2));
% Separate obstacles to large and small ones + filter them
gtl = filter_obstacles(gtl, eval_params.img_size, tmp_inv_sea_mask);
%% Rectify image if it is needed
% remap segmentation mask to rectified images if we are evaluation
% on the rectified dataset and if the segmentation masks were
% obtained on raw images
if(eval_params.rectified == 1 && seq.is_rectified == 0)
msk = imresize(cv.remap(imresize(msk, eval_params.img_size, 'Method', 'nearest'), map_L1, map_L2), [msk_size_y, msk_size_x], 'Method', 'nearest');
end
%{
if(eval_params.rectified == 1)
% remove black areas
for rgb_counter = 1 : 3
tmp = msk(:, :, rgb_counter); % extract color
tmp(black_areas == 1) = eval_params.labels(3, rgb_counter);
msk(:, :, rgb_counter) = tmp;
end
end
%}
%% Postprocess output segmentation images
% Get extracted sea-mask and a list of all detections
[det_objs, sea_mask] = postprocess_output_image(msk, eval_params);
output_detections{counter, 1} = logical(sea_mask);
output_detections{counter, 2} = det_objs;
%% Perform evaluation of detections...
[rmse_water, tp, fp, fn, ~, ~, ~] = evaluate_detections_modd2(sea_mask, det_objs, gtl, eval_params);
output_results_cell{counter, 1} = rmse_water;
output_results_cell{counter, 2} = tp;
output_results_cell{counter, 3} = fp;
output_results_cell{counter, 4} = fn;
counter = counter + 1;
end
close(f);
end