arron2003 / rgbd2full3d

Matlab codes that convert a RGBD image into a cad like model. This code is released as a part of my PhD dissertation.
Other
24 stars 13 forks source link

Error running demo.m #1

Closed mrkulk closed 10 years ago

mrkulk commented 10 years ago

I get the following error running the code. --

Preprocessing RGBD data.. Computing Normal.. Computing Room Rotation.. Parsing with NYU segmentation.. Extracting SIFT.. Fitting planes.. Computing PB.. Merging Level 1..Error using pca (line 324) The estimated covariance matrix is not positive semi-definite. Consider using 'complete' instead of 'pairwise' option.

Error in princomp (line 26) varargout{1} = pca(varargin{1},'Algorithm','eig','Economy',fEconomy);

Error in fit_plane_pca (line 10) coeff = princomp(points3d);

Error in NYUparse>get_features_3d (line 307) surfaceNormals(ii,:) = fit_plane_pca(samplePoints{ii});

Error in NYUparse (line 60) features3d = get_features_3d(boundaryInfo, featureData, points3d, normals, regionInteriors);

Error in demo (line 23) load(['cache/' opt.name '.mat']); data=NYUparse(data, opt);

arron2003 commented 10 years ago

Sorry for the very late reply.

I believe they stem from the changed behavior of new versions of matlab and it will be trivial to fix them (I was running 2009b). For PCA warnings, you can change from princomp(...) to "pca(...,'Algorithm','svd','Economy',false)" For ismember warnings, you can change to "ismember(..., 'legacy')"

Unfortunately I do not own a newer version of matlab to fully test it. I shall fix these issue once I have access to newer version of matlab.

mrkulk commented 10 years ago

Yes I tested that the same day and everything works fine! Thanks...

jiunyen-ching commented 3 years ago

Hi @arron2003 @mrkulk, do you still have the files downloaded via setup.m? I tried downloading them but the links are now broken.

arron2003 commented 3 years ago

You can download it at https://dhoiem.cs.illinois.edu/publications/iccv13-cleaned_3D.zip

jiunyen-ching commented 3 years ago

Thanks for the new link @arron2003!

I tried running pipeline_preprocess.m by placing nyu_depth_v2_labeled.mat and support_labels.mat downloaded from

http://horatio.cs.nyu.edu/mit/silberman/indoor_seg_sup/

as such

image

but I ran into an error saying that Subscripted assignment dimension mismatch.

Full stack trace: >> pipeline_preprocess The size of scenes = 464 Subscripted assignment dimension mismatch.

Error in processRoom (line 8) d(opt.mask(1):opt.mask(3),opt.mask(2):opt.mask(4)) = data.depths;

Error in pipeline_preprocess (line 79) data = processRoom(data);

I tried changing dimensions of d in processRoom.m from d = zeros(opt.orig_H, opt.orig_W); to d = zeros(opt.H, opt.W); to match the dimensions of depths variable in t struct but I still get the same error. Is there any other changes that I could try?

jiunyen-ching commented 3 years ago

Update: Probably due to some weird bug in Windows, but problem went away after arranging extracted folder like so /Downloads/cleaned_3D instead of /Downloads/iccv13-cleaned_3D/cleaned_3D.