----------------------------------------------------------------------------

Estimating the 3D Layout of Indoor Scenes and its Clutter from Depth Sensors

----------------------------------------------------------------------------
1. General description of data and folders

(1) ./dcBP/ contains the underlying inference engine of our method. For more de-
tails, please refer to [1].

(2) ./label/ contains the the annotation and other groundtruth information 
of the 303 samples.

(3) ./Miscellaneous/ contains auxiliary data to support the running of inference 
and visualization.

(4) ./pattern/ contains the features for 303 samples. For more details, please
refer to [2]. 

(5) ./Utility/ contains functions for inference and visualization.

(6) dataset_split.mat is the splitting of training and test set.

(7) err_ICCV_C_5000.mat is the layout and labeling accuracy on our test set samples.

(8) param_ICCV_C_5000.mat is the configuration for running various scripts.

(9) results_ICCV_C_5000.mat is the variable values as the result of layout 
and labeling estimations on the test set.

(10) w_ICCV_C_5000.mat is the weight of our model described in [2].  



2. Dependency

To activate the dcBP inference engine, a MPI implementation needs to be installed.
eg. MPICH2 v1.4.1 is the one we use to test the code here.



3. Top-level Script

(1) GetIntegralCompatibilityFast.m generates Integral Compatability with the fast 
method described in [2]. 

(2) InferTestset.m is to do inference on the test set with given model 
w_ICCV_C_5000.mat. Line 18 should be updated with a valid MPI excutable file,
e.g. MPICH2_1.4.1\bin\mpiexec.exe.

(3) VisualizeClutterSemantic.m can be used to generate the clutter semantic 
visualization in [2]. Line 20 needs to be update with the directory of original
NYU depth V2 images. You might also need to replace function GetImage with 
your own implementation.

(4) VisualizeLabelingEstimation.m can visualize the 6-class labeling estimation.

(5) VisualizeLayoutEstimation.m produces visualization of the layout estimation
together with the groundtruth. Line 10 need to be update with the directory of 
original NYU depth V2 images. Function GetImage in Line 32 should also be replaced.



4. Reference

[1] A.G. Schwing, T. Hazan, M. Pollefeys and R. Urtasun; 
Distributed Message Passing for Large Scale Graphical Models; IEEE Conf. on 
Computer Vision and Pattern Recognition (CVPR); 2011

[2] J. Zhang, C. Kan, A.G. Schwing and R. Urtasun; 
Estimating the 3D Layout of Indoor Scenes and its Clutter from Depth Sensors; 
International Conference on Computer Vision (ICCV); 2013