1 | % * This code was used in the following articles:
|
---|
2 | % * [1] Learning 3-D Scene Structure from a Single Still Image,
|
---|
3 | % * Ashutosh Saxena, Min Sun, Andrew Y. Ng,
|
---|
4 | % * In ICCV workshop on 3D Representation for Recognition (3dRR-07), 2007.
|
---|
5 | % * (best paper)
|
---|
6 | % * [2] 3-D Reconstruction from Sparse Views using Monocular Vision,
|
---|
7 | % * Ashutosh Saxena, Min Sun, Andrew Y. Ng,
|
---|
8 | % * In ICCV workshop on Virtual Representations and Modeling
|
---|
9 | % * of Large-scale environments (VRML), 2007.
|
---|
10 | % * [3] 3-D Depth Reconstruction from a Single Still Image,
|
---|
11 | % * Ashutosh Saxena, Sung H. Chung, Andrew Y. Ng.
|
---|
12 | % * International Journal of Computer Vision (IJCV), Aug 2007.
|
---|
13 | % * [6] Learning Depth from Single Monocular Images,
|
---|
14 | % * Ashutosh Saxena, Sung H. Chung, Andrew Y. Ng.
|
---|
15 | % * In Neural Information Processing Systems (NIPS) 18, 2005.
|
---|
16 | % *
|
---|
17 | % * These articles are available at:
|
---|
18 | % * http://make3d.stanford.edu/publications
|
---|
19 | % *
|
---|
20 | % * We request that you cite the papers [1], [3] and [6] in any of
|
---|
21 | % * your reports that uses this code.
|
---|
22 | % * Further, if you use the code in image3dstiching/ (multiple image version),
|
---|
23 | % * then please cite [2].
|
---|
24 | % *
|
---|
25 | % * If you use the code in third_party/, then PLEASE CITE and follow the
|
---|
26 | % * LICENSE OF THE CORRESPONDING THIRD PARTY CODE.
|
---|
27 | % *
|
---|
28 | % * Finally, this code is for non-commercial use only. For further
|
---|
29 | % * information and to obtain a copy of the license, see
|
---|
30 | % *
|
---|
31 | % * http://make3d.stanford.edu/publications/code
|
---|
32 | % *
|
---|
33 | % * Also, the software distributed under the License is distributed on an
|
---|
34 | % * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either
|
---|
35 | % * express or implied. See the License for the specific language governing
|
---|
36 | % * permissions and limitations under the License.
|
---|
37 | % *
|
---|
38 | % */
|
---|
39 | function []=gen_feature_filt1D_sepH2H4_fixMem(batchNumber,HistFeaType,Absolute) |
---|
40 | |
---|
41 | % This function calculate the feature of each subsuperpixel using texture |
---|
42 | % infomation |
---|
43 | |
---|
44 | % decide the Hist |
---|
45 | if strcmp(HistFeaType,'Whole') |
---|
46 | Hist = 1; |
---|
47 | HistFeaType |
---|
48 | else |
---|
49 | Hist = 0; |
---|
50 | end |
---|
51 | |
---|
52 | if Hist ~= 1 && Absolute ~=1 |
---|
53 | return |
---|
54 | end |
---|
55 | |
---|
56 | if nargin < 1 |
---|
57 | batchNumber = 1; |
---|
58 | elseif nargin < 2 |
---|
59 | Hist = 1; |
---|
60 | Absolute =1; |
---|
61 | elseif nargin < 3 |
---|
62 | Absolute =1; |
---|
63 | end |
---|
64 | |
---|
65 | global GeneralDataFolder ScratchDataFolder LocalFolder ClusterExecutionDirectory... |
---|
66 | ImgFolder VertYNuPatch VertYNuDepth HoriXNuPatch HoriXNuDepth a_default b_default Ox_default Oy_default... |
---|
67 | Horizon_default filename batchSize NuRow_default SegVertYSize SegHoriXSize WeiBatchSize PopUpVertY PopUpHoriX taskName... |
---|
68 | TrainVerYSize TrainHoriXSize MempryFactor; |
---|
69 | |
---|
70 | |
---|
71 | % load estimated sky |
---|
72 | load([ScratchDataFolder '/data/MaskGSky.mat']); % maskg is the estimated ground maskSky is the estimated sky |
---|
73 | |
---|
74 | % load([ScratchDataFolder '/data/filename.mat']);% load the filename |
---|
75 | % load([ScratchDataFolder '/data/PlaneParameterTure.mat']); % planeParameter |
---|
76 | load([ScratchDataFolder '/data/LowResImgIndexSuperpixelSep.mat']); % superpixel_index |
---|
77 | %load([ScratchDataFolder '/data/MediResImgIndexSuperpixelSep.mat']); % MediResImgIndexSuperpixelSep |
---|
78 | load([ScratchDataFolder '/data/DiffLowResImgIndexSuperpixelSep.mat']); % DiffLowResImgIndexSuperpixelSep |
---|
79 | DiffLowResImgIndexSuperpixelSep = DiffLowResImgIndexSuperpixelSep(:,1);% need only the middle scale segmentation |
---|
80 | |
---|
81 | %load([ScratchDataFolder '/data/FeatureSuperpixel.mat']); %load feature of superpixel |
---|
82 | % load FeaMax to do normalizeing |
---|
83 | load([GeneralDataFolder '/FeaMax.mat']); |
---|
84 | FeaMax = 10.^floor(log10(FeaMax)); |
---|
85 | |
---|
86 | % prepare data step |
---|
87 | nu_pics = size(filename,2); % number of pictures |
---|
88 | |
---|
89 | |
---|
90 | % ====================== change able parameter ===================== |
---|
91 | %batchSize = 10;% decide the batch size as 10 image pre batch |
---|
92 | % ================================================================== |
---|
93 | |
---|
94 | % for batchImg = 1:batchSize:nu_pics |
---|
95 | batchImg = 1:batchSize:nu_pics; |
---|
96 | |
---|
97 | f = []; % total feature: feature of superpixel followed by texture feature of patch |
---|
98 | PicsNu = 1 |
---|
99 | for i = batchImg(batchNumber):min(batchImg(batchNumber)+batchSize-1, nu_pics) |
---|
100 | i |
---|
101 | % load MediResImgIndexSuperpixelSep |
---|
102 | load([ScratchDataFolder '/data/MedSeg/MediResImgIndexSuperpixelSep' num2str(i) '.mat']); |
---|
103 | |
---|
104 | % calculate all the features, ray, plane parameter, and row column value |
---|
105 | img = imread([GeneralDataFolder '/' ImgFolder '/' filename{i} '.jpg']);% read in the hi resolution image of the ith file |
---|
106 | size(img) |
---|
107 | % change the resolution in to exactly [TrainVerYSize TrainHoriXSize] |
---|
108 | |
---|
109 | if ~all(size(img) == [TrainVerYSize TrainHoriXSize 3]) |
---|
110 | disp('resize to 2272 1704') |
---|
111 | img = imresize(img,[TrainVerYSize TrainHoriXSize],'bilinear'); |
---|
112 | end |
---|
113 | % the images resolusion must be bigger then a certain size to have reasonable predicted depth |
---|
114 | % if prod(size(img))<SegVertYSize*SegHoriXSize*3 |
---|
115 | % img = imresize(img,[SegVertYSize SegHoriXSize],'bilinear'); |
---|
116 | % the images resolusion must be smaller then a certain size to avoid out of memory |
---|
117 | % elseif prod(size(img)) > TrainVerYSize*TrainHoriXSize*3*(MempryFactor); |
---|
118 | % disp('origin size') |
---|
119 | % size(img) |
---|
120 | % img = imresize(img,[TrainVerYSize TrainHoriXSize],'bilinear'); |
---|
121 | % disp('image too big') |
---|
122 | % size(img) |
---|
123 | % end |
---|
124 | [vertical_size_hi_res horizontal_size_hi_res t] = size(img); clear t; |
---|
125 | % get the horizontal(vertical_size_hi_res) and vertical(horizontal_size_hi_res) size of hi resolution image |
---|
126 | sup_hi_res = imresize(MediResImgIndexSuperpixelSep, ... |
---|
127 | [vertical_size_hi_res horizontal_size_hi_res],'nearest');% enlarge the low res superpixel into hi res superpixel |
---|
128 | clear MediResImgIndexSuperpixelSep; |
---|
129 | |
---|
130 | % generate the superpixel in the depth_grid size |
---|
131 | sup = imresize(LowResImgIndexSuperpixelSep{i,1},[VertYNuDepth HoriXNuDepth],'nearest'); |
---|
132 | |
---|
133 | % load picsinfo just for the horizontal value |
---|
134 | PicsinfoName = strrep(filename{i},'img','picsinfo'); |
---|
135 | temp = dir([GeneralDataFolder '/PicsInfo/' PicsinfoName '.mat']); |
---|
136 | if size(temp,1) == 0 |
---|
137 | a = a_default; |
---|
138 | b = b_default; |
---|
139 | Ox = Ox_default; |
---|
140 | Oy = Oy_default; |
---|
141 | Horizon = Horizon_default; |
---|
142 | else |
---|
143 | load([GeneralDataFolder '/PicsInfo/' PicsinfoName '.mat']); |
---|
144 | end |
---|
145 | |
---|
146 | % calculate the ray |
---|
147 | RayCenter = GenerateRay(HoriXNuDepth,VertYNuDepth,'center',a,b,Ox,Oy); %[ horiXSizeLowREs VertYSizeLowREs 3] |
---|
148 | |
---|
149 | % calculate how many mask we need |
---|
150 | NuMask = ceil([HoriXNuDepth/HoriXNuPatch; VertYNuDepth/VertYNuPatch]) |
---|
151 | |
---|
152 | % Grid Info |
---|
153 | gridinfo = [horizontal_size_hi_res HoriXNuDepth HoriXNuPatch; vertical_size_hi_res VertYNuDepth VertYNuPatch]; |
---|
154 | ratio(1:2) = gridinfo(:,1)./gridinfo(:,end); |
---|
155 | ratio(3:4) = gridinfo(:,1)./gridinfo(:,2); |
---|
156 | |
---|
157 | % calcuate the position of the mask |
---|
158 | hight(1) = round((floor(ratio(2))-1)/2); |
---|
159 | hight(2) = floor(ratio(2)) - 1 - hight(1); |
---|
160 | width(1) = round((floor(ratio(1))-1)/2); |
---|
161 | width(2) = floor(ratio(1)) - 1 - width(1); |
---|
162 | |
---|
163 | %big_sup_depthmap_res = imresize(LowResImgIndexSuperpixelSep{i,2}, ... |
---|
164 | % [VertYNuDepth HoriXNuDepth],'nearest');% enlarge the low res superpixel into hi res superpixel |
---|
165 | f_pics = []; |
---|
166 | tic |
---|
167 | [H2] = calculateFilterBanks_old(img); % (hard work 1min) use Ashutaosh's code |
---|
168 | H2 = H2.^2; |
---|
169 | % NuFeaH2H4 = 2*size(H2,3); |
---|
170 | hcol = ones(floor(ratio(2)),1); |
---|
171 | hrow = ones(1,floor(ratio(1))); |
---|
172 | hrect = ones(floor(ratio(2)),floor(ratio(1))); |
---|
173 | size(hcol) |
---|
174 | size(hrow) |
---|
175 | size(hrect) |
---|
176 | |
---|
177 | if Absolute == 1 |
---|
178 | % highest resolution |
---|
179 | row_start = 1; |
---|
180 | % MinTest = zeros(55*305,1); |
---|
181 | for j = 1:NuMask(1) |
---|
182 | for k = 1:NuMask(2); |
---|
183 | f_pics_mask = []; |
---|
184 | % first generate the mask respect to the dominate |
---|
185 | % subsuperpixel |
---|
186 | [mask,SupIndex,PixelMask,PatchMask] = makeSubSupMask(gridinfo,sup_hi_res,sup,[j; k],width,hight); |
---|
187 | % calculaing the normalize value |
---|
188 | NormalizeValue = conv2(hcol,hrow,mask,'same'); |
---|
189 | |
---|
190 | % 1) zeroth feature the superpixel index to keep a record |
---|
191 | f_pics_mask = [f_pics_mask SupIndex]; |
---|
192 | clear SupIndex; |
---|
193 | |
---|
194 | % 2) generate the 1:34 features for H2 for 1 center and 4 neighbor (left right top bottom) |
---|
195 | fInd=2; |
---|
196 | for m = 1:17 |
---|
197 | %temp1 = conv2(hcol,hrow,H2(:,:,m).*(mask),'same'); |
---|
198 | temp = conv2(H2(:,:,m).*(mask),hrect,'same'); |
---|
199 | size(temp) |
---|
200 | f_pics_mask = [f_pics_mask temp(PixelMask)./NormalizeValue(PixelMask)... |
---|
201 | ./FeaMax(1,fInd)]; |
---|
202 | fInd = fInd+1; |
---|
203 | end |
---|
204 | |
---|
205 | f_pics(PatchMask, row_start:row_start+size(f_pics_mask,2)-1) = f_pics_mask; |
---|
206 | |
---|
207 | end |
---|
208 | end |
---|
209 | |
---|
210 | H2 = H2.^2; % This H2 is H4 |
---|
211 | row_start = row_start+size(f_pics_mask,2); |
---|
212 | |
---|
213 | % RAJIV MIN -- ERROR ------- this part which sums up the filter response in a superpixel, should |
---|
214 | % be done using integral images, prefably in C++ code. |
---|
215 | |
---|
216 | for j = 1:NuMask(1) |
---|
217 | for k = 1:NuMask(2); |
---|
218 | f_pics_mask = []; |
---|
219 | % first generate the mask respect to the dominate |
---|
220 | % subsuperpixel |
---|
221 | [mask,SupIndex,PixelMask,PatchMask] = makeSubSupMask(gridinfo,sup_hi_res,sup,[j; k],width,hight); |
---|
222 | % calculaing the normalize value |
---|
223 | NormalizeValue = conv2(hcol,hrow,mask,'same'); |
---|
224 | |
---|
225 | % 1) zeroth feature the superpixel index to keep a record |
---|
226 | % f_pics_mask = [f_pics_mask SupIndex]; |
---|
227 | clear SupIndex; |
---|
228 | |
---|
229 | % 2) generate the 1:34 features for H4 for 1 center and 4 neighbor (left right top bottom) |
---|
230 | fInd=19; |
---|
231 | for m = 1:17 |
---|
232 | %temp = conv2(hcol,hrow,H2(:,:,m).*(mask),'same'); |
---|
233 | temp = conv2(H2(:,:,m).*(mask),hrect,'same'); |
---|
234 | f_pics_mask = [f_pics_mask temp(PixelMask)./NormalizeValue(PixelMask)... |
---|
235 | ./FeaMax(1,fInd)]; |
---|
236 | fInd = fInd+1; |
---|
237 | end |
---|
238 | |
---|
239 | f_pics(PatchMask, row_start:row_start+size(f_pics_mask,2)-1) = f_pics_mask; |
---|
240 | |
---|
241 | end |
---|
242 | end |
---|
243 | end |
---|
244 | |
---|
245 | if Hist == 1 |
---|
246 | % RAJIV ---- ERROR -- should be disabled. |
---|
247 | H2 = H2.^(0.5); |
---|
248 | % =============calculate the histagram of the features for relative depth estimation================ |
---|
249 | disp('cal_relative') |
---|
250 | [relativeFeatureVector] = makeRelativeFeatureVector(H2,1); |
---|
251 | % ================================================================================================== |
---|
252 | end |
---|
253 | clear H2; |
---|
254 | |
---|
255 | % 1/3 resolution |
---|
256 | feaScale1 = []; |
---|
257 | MedResY = round(gridinfo(2,1)/3); |
---|
258 | MedResX = round(gridinfo(1,1)/3); |
---|
259 | DepthGridSizeY = MedResY/VertYNuDepth; |
---|
260 | DepthGridSizeX= MedResX/HoriXNuDepth; |
---|
261 | imgMedRes = imresize(img,[MedResY MedResX],'nearest'); |
---|
262 | clear img; |
---|
263 | [H2] = calculateFilterBanks_old(imgMedRes); % (hard work 1min) use Ashutaosh's code |
---|
264 | H2 = H2.^2; |
---|
265 | H4 = H2.^2; |
---|
266 | if Absolute == 1 |
---|
267 | row_start = size(f_pics,2)+1; |
---|
268 | % calculating number of mask |
---|
269 | NormalizeValue = conv2(hcol,hrow,ones(MedResY,MedResX),'same'); |
---|
270 | NuMask = ceil([HoriXNuDepth/HoriXNuPatch*3; VertYNuDepth/VertYNuPatch*3]); |
---|
271 | |
---|
272 | % 1) generating the PixelMask |
---|
273 | PixelMask = logical(zeros(MedResY,MedResX)); |
---|
274 | [X Y] = meshgrid(ceil((1/2)*DepthGridSizeX:DepthGridSizeX:MedResX),... |
---|
275 | ceil((1/2)*DepthGridSizeY:DepthGridSizeY:MedResY)); |
---|
276 | PixelMask = sub2ind(size(PixelMask),Y(:),X(:)); |
---|
277 | for m = 1:17 |
---|
278 | temp = conv2(hcol,hrow,H2(:,:,m),'same');% 102.533776 seconds |
---|
279 | feaScale1(:,m) = temp(PixelMask)./NormalizeValue(PixelMask)... |
---|
280 | ./FeaMax(1,fInd); |
---|
281 | fInd = fInd +1; |
---|
282 | end |
---|
283 | |
---|
284 | for m = 1:17 |
---|
285 | temp = conv2(hcol,hrow,H4(:,:,m),'same');% 102.533776 seconds |
---|
286 | feaScale1(:,m+17) = temp(PixelMask)./NormalizeValue(PixelMask)... |
---|
287 | ./FeaMax(1,fInd); |
---|
288 | fInd = fInd +1; |
---|
289 | end |
---|
290 | |
---|
291 | % shift = [0 0; -1 0; 1 0; 0 -1; 0 1].*repmat(NuMask',[5 1]); |
---|
292 | % for l = 1:5 |
---|
293 | % [Ix Iy] = meshgrid(max(min(2+shift(l,1):HoriXNuDepth+2-1+shift(l,1),HoriXNuDepth+2),1),... |
---|
294 | % max(min(2+shift(l,2):VertYNuDepth+2-1+shift(l,2),VertYNuDepth+2),1)); |
---|
295 | % maskNeibor = sub2ind([VertYNuDepth+2, HoriXNuDepth+2], Iy(:), Ix(:)); |
---|
296 | % f_pics = [f_pics feaScale1(maskNeibor,:)]; |
---|
297 | % end |
---|
298 | |
---|
299 | size(f_pics) |
---|
300 | size(feaScale1) |
---|
301 | f_pics = [f_pics feaScale1]; |
---|
302 | end |
---|
303 | clear H4; |
---|
304 | |
---|
305 | if Hist == 1 |
---|
306 | % =============calculate the histagram of the features for relative depth estimation================ |
---|
307 | [relativeFeatureVector] = cat(3,relativeFeatureVector,makeRelativeFeatureVector(H2,2)); |
---|
308 | % ================================================================================================== |
---|
309 | end |
---|
310 | |
---|
311 | clear H2; |
---|
312 | |
---|
313 | % 1/9 resolution |
---|
314 | feaScale1 = []; |
---|
315 | LowResY = round(gridinfo(2,1)/9); |
---|
316 | LowResX = round(gridinfo(1,1)/9); |
---|
317 | DepthGridSizeY = LowResY/VertYNuDepth; |
---|
318 | DepthGridSizeX= LowResX/HoriXNuDepth; |
---|
319 | imgLowRes = imresize(imgMedRes,[LowResY LowResX],'nearest'); |
---|
320 | clear imgMedRes; |
---|
321 | [H2] = calculateFilterBanks_old(imgLowRes); % (hard work 1min) use Ashutaosh's code |
---|
322 | H2 = H2.^2; |
---|
323 | H4 = H2.^2; |
---|
324 | |
---|
325 | if Absolute == 1 |
---|
326 | row_start = size(f_pics,2)+1; |
---|
327 | % calculating number of mask |
---|
328 | NormalizeValue = conv2(hcol,hrow,ones(LowResY,LowResX),'same'); |
---|
329 | NuMask = ceil([HoriXNuDepth/HoriXNuPatch*9; VertYNuDepth/VertYNuPatch*9]); |
---|
330 | |
---|
331 | % 1) generating the PixelMask |
---|
332 | PixelMask = logical(zeros(LowResY,LowResX)); |
---|
333 | [X Y] = meshgrid(ceil((1/2)*DepthGridSizeX:DepthGridSizeX:LowResX),... |
---|
334 | ceil((1/2)*DepthGridSizeY:DepthGridSizeY:LowResY)); |
---|
335 | PixelMask = sub2ind(size(PixelMask),Y(:),X(:)); |
---|
336 | % size(maskInd) |
---|
337 | % PixelMask(maskInd) = true; |
---|
338 | %PixelMask = logical(zeros(LowResY,LowResX)); |
---|
339 | %PixelMask(ceil((1/2)*DepthGridSizeY:DepthGridSizeY:LowResY),... |
---|
340 | % ceil((1/2)*DepthGridSizeX:DepthGridSizeX:LowResX)) = true; |
---|
341 | % valuemask = sum(PixelMask,1)>0; |
---|
342 | % if all(PixelMask(1,valuemask) == true) |
---|
343 | % PixelMask(2,valuemask) = true; |
---|
344 | % else |
---|
345 | % PixelMask(1,valuemask) = true; |
---|
346 | % end |
---|
347 | % if all(PixelMask(end,valuemask) == true) |
---|
348 | % PixelMask(end-1,valuemask) = true; |
---|
349 | % else |
---|
350 | % PixelMask(end,valuemask) = true; |
---|
351 | % end |
---|
352 | % valuemask = sum(PixelMask,2)>0; |
---|
353 | % if all(PixelMask(valuemask,1) == true) |
---|
354 | % PixelMask(valuemask,2) = true; |
---|
355 | % else |
---|
356 | % PixelMask(valuemask,1) = true; |
---|
357 | % end |
---|
358 | % if all(PixelMask(valuemask,end) == true); |
---|
359 | % PixelMask(valuemask,end-1) = true; |
---|
360 | % else |
---|
361 | % PixelMask(valuemask,end) = true; |
---|
362 | % end |
---|
363 | |
---|
364 | % 2) generate the 1:34 features for H2 and H4 for 1 center and 4 neighbor (left right top bottom) |
---|
365 | for m = 1:17 |
---|
366 | temp = conv2(hcol,hrow,H2(:,:,m),'same');% 102.533776 seconds |
---|
367 | feaScale1(:,m) = temp(PixelMask)./NormalizeValue(PixelMask)... |
---|
368 | ./FeaMax(1,fInd); |
---|
369 | fInd = fInd +1; |
---|
370 | end |
---|
371 | |
---|
372 | for m = 1:17 |
---|
373 | temp = conv2(hcol,hrow,H4(:,:,m),'same');% 102.533776 seconds |
---|
374 | feaScale1(:,m+17) = temp(PixelMask)./NormalizeValue(PixelMask)... |
---|
375 | ./FeaMax(1,fInd); |
---|
376 | fInd = fInd +1; |
---|
377 | end |
---|
378 | |
---|
379 | % shift = [0 0; -1 0; 1 0; 0 -1; 0 1].*repmat(NuMask',[5 1]); |
---|
380 | % for l = 1:5 |
---|
381 | % [Ix Iy] = meshgrid(max(min(2+shift(l,1):HoriXNuDepth+2-1+shift(l,1),HoriXNuDepth+2),1),... |
---|
382 | %% max(min(2+shift(l,2):VertYNuDepth+2-1+shift(l,2),VertYNuDepth+2),1)); |
---|
383 | % maskNeibor = sub2ind([VertYNuDepth+2, HoriXNuDepth+2], Iy(:), Ix(:)); |
---|
384 | % f_pics = [f_pics feaScale1(maskNeibor,:)]; |
---|
385 | % end |
---|
386 | f_pics = [f_pics feaScale1]; |
---|
387 | end |
---|
388 | clear H4; |
---|
389 | |
---|
390 | if Hist ==1 |
---|
391 | % =============calculate the histagram of the features for relative depth estimation================ |
---|
392 | [relativeFeatureVector] = cat(3,relativeFeatureVector,makeRelativeFeatureVector(H2,5)); |
---|
393 | RFVector{PicsNu} = relativeFeatureVector; |
---|
394 | clear relativeFeatureVector; |
---|
395 | DateStamp = date; |
---|
396 | save([ScratchDataFolder '/data/feature_Hist_Whole' num2str(batchNumber) '_' DateStamp '.mat'],'RFVector'); |
---|
397 | % ================================================================================================== |
---|
398 | end |
---|
399 | |
---|
400 | clear H2; |
---|
401 | |
---|
402 | if Absolute == 1 |
---|
403 | % superpixel features |
---|
404 | %fsup = FeatureSuperpixel{i}; |
---|
405 | %f_pics = [f_pics fsup(:,f_pics(:,1))']; |
---|
406 | % other features without relation with neiborfeatures |
---|
407 | % 1) the closest ground position to the (i, j) patch |
---|
408 | % How can we remove DiffLowResImgIndexSuperpixelSep since generate another superpixel takes time |
---|
409 | big_sup = imresize(DiffLowResImgIndexSuperpixelSep{i,1},[VertYNuDepth HoriXNuDepth],'nearest'); |
---|
410 | GroundSupIndex = unique(big_sup(maskg{i})); |
---|
411 | |
---|
412 | % NonGround = []; |
---|
413 | % for j = 1:HoriXNuDepth |
---|
414 | % Ground(j) = analysesupinpatch(big_sup(round(gridinfo(2,2)*(1-Horizon)):end,j)); |
---|
415 | % NonGround(j) = analysesupinpatch(big_sup(round(1:(gridinfo(2,2)*(1-Horizon)-1)),j)); |
---|
416 | % NonGround = [NonGround (unique(big_sup(round(1:(gridinfo(2,2)*(1-Horizon)-1)),j)))']; |
---|
417 | % end |
---|
418 | % Ground = setdiff(Ground,NonGround); |
---|
419 | for j = 1:HoriXNuDepth |
---|
420 | Gmask = logical(zeros(size( big_sup(:,j)))); |
---|
421 | for k=GroundSupIndex' |
---|
422 | Gmask(big_sup(:,j) == k) = true; |
---|
423 | end |
---|
424 | [rowSub colSub ] = find(Gmask); |
---|
425 | minRow = min(rowSub); |
---|
426 | % if size(minRow,1) == 0 |
---|
427 | % Gmask = logical(zeros(size( big_sup(:,j)))); |
---|
428 | % MaybeGround = unique(big_sup(round(gridinfo(2,2)*(1-Horizon)):end,j)); |
---|
429 | % MaybeGround = setdiff(MaybeGround,NonGround); |
---|
430 | % for k=MaybeGround |
---|
431 | % Gmask(big_sup(:,j) == k) = true; |
---|
432 | % end |
---|
433 | % [rowSub colSub ] = find(Gmask); |
---|
434 | % minRow = min(rowSub); |
---|
435 | % if size(minRow,1) == 0 |
---|
436 | % lastguessGround = analysesupinpatch(big_sup(round(gridinfo(2,2)*(1-Horizon)):end,j)); |
---|
437 | % [rowSub colSub ] = find(big_sup(round(gridinfo(2,2)*(1-Horizon)):end,j)==lastguessGround); |
---|
438 | % minRow = min(rowSub); |
---|
439 | % end |
---|
440 | % end |
---|
441 | %minRow = max(minRow,round(gridinfo(2,2)*(1-Horizon))); |
---|
442 | |
---|
443 | GroundVertEdge(:,j) = (1-(VertYNuDepth-1)/VertYNuDepth)./RayCenter(:,j,3); |
---|
444 | if size(minRow,1) ~= 0 |
---|
445 | GroundVertEdge((minRow+1):VertYNuDepth,j) = (1-((minRow+1):VertYNuDepth)'/VertYNuDepth)./RayCenter((minRow+1):VertYNuDepth,j,3); |
---|
446 | GroundVertEdge(1:(minRow),j) = (1- minRow/VertYNuDepth)./RayCenter(1:(minRow),j,3); |
---|
447 | end |
---|
448 | end |
---|
449 | GroundVertEdgePics = GroundVertEdge; |
---|
450 | |
---|
451 | |
---|
452 | %f_pics(:,end+1) = max([GroundVertEdgePics(:) repmat((1:VertYNuDepth)',[HoriXNuDepth 1])],[],2); |
---|
453 | %f_pics(:,end+1) = Gr undVertEdgePics(:);%./reshape(RayCenter(:,:,3),[],1); |
---|
454 | f_pics(:,end+1) = min([GroundVertEdgePics(:)... |
---|
455 | repmat((VertYNuDepth:-1:1)'/VertYNuDepth,[HoriXNuDepth 1])./reshape(RayCenter(:,:,3),[],1)],[],2); |
---|
456 | |
---|
457 | disp(['Image Number ' num2str(i)]); |
---|
458 | f{PicsNu} = f_pics; |
---|
459 | % Absolute Features are index by Type BatchNu Data |
---|
460 | DateStamp = date; |
---|
461 | save([ScratchDataFolder '/data/feature_Abs_Whole' num2str(batchNumber) '_' DateStamp '.mat'],'f'); |
---|
462 | |
---|
463 | toc; |
---|
464 | %return; |
---|
465 | end |
---|
466 | PicsNu = PicsNu + 1 |
---|
467 | batchNumber |
---|
468 | end |
---|
469 | % Absolute Features are index by Type BatchNu Data |
---|
470 | DateStamp = date; |
---|
471 | if Absolute == 1 |
---|
472 | save([ScratchDataFolder '/data/feature_Abs_Whole' num2str(batchNumber) '_' DateStamp '.mat'],'f'); |
---|
473 | end |
---|
474 | % Hist Features are index by Type BatchNu Data |
---|
475 | if Hist == 1 |
---|
476 | save([ScratchDataFolder '/data/feature_Hist_Whole' num2str(batchNumber) '_' DateStamp '.mat'],'RFVector'); |
---|
477 | end |
---|
478 | return; |
---|