Error using UGM_makeEdgeVEC

[TJ1451154]

hello, Thank you for sharing your code.
my operating system is ubuntu16.04. I downloaded UGM_2011.zip and compiled it successfully.
when i run benchmark.m:
Error using UGM_makeEdgeVEC
Requested 4296291607x7311422349372966656 (17179869184.0GB) array exceeds maximum array size preference. Creation of arrays greater than this limit may take a long time and
cause MATLAB to become unresponsive. See array size limit or preference panel for more information.
Error in UGM_makeEdgeVE (line 11)
[V,E] = UGM_makeEdgeVEC(int32(edgeEnds),int32(nNodes));

Error in UGM_makeEdgeStruct (line 26)
[V,E] = UGM_makeEdgeVE(edgeEnds,nNodes,useMex);

Error in LBP_sum_product (line 21)
edgeStruct = UGM_makeEdgeStruct(adjacency,nClasses,1);

Error in benchmark (line 37)
[p_lpb_sp, T] = LBP_sum_product(initial_classif, graph, .5, 10);

[loicland]

Could you show the values of int32(edgeEnds) and int32(nNodes)?

[TJ1451154]

Could you show the values of int32(edgeEnds) and int32(nNodes)?
when run LBP sum product method.
input 1324310
[V,E] = UGM_makeEdgeVE(edgeEnds,nNodes,useMex)
edgeEnds=7291304X2 int32 matrix
nNodes=1324310 int32
I changed the input and reduced the number of input points.
input 100000
[V,E] = UGM_makeEdgeVE(edgeEnds,nNodes,useMex)
edgeEnds 544574X2 int32 matrix
nNodes 100000 int32
But the program also reports an error:
Error using UGM_makeEdgeVEC
Requested 4296291607x7903554357059420160 (17179869184.0GB) array exceeds maximum array size preference. Creation of arrays greater than this limit may take a long time and
cause MATLAB to become unresponsive. See array size limit or preference panel for more information.

[loicland]

Please note that UGM is not my code.

Requested 4296291607x7903554357059420160 array

clearly your inputs are wrong. Set useMex to 0 in the call to UGM_makeEdgeStruct to have a more useful error.

Further questions: are you working on the Oakland dataset? Could you write the size and first values of graph.source and graph.target?

[TJ1451154]

Thank you very much for your help.
after running the following code:
graph = build_graph_structure('oakland.ply',10,0,0);
load('oakland_RF', 'initial_classif');
load('oakland_GT', 'ground_truth'); %MUST BE UINT8
the values are as follows:
graph.XYZ= 1324310 _x_3 single graph.source=1458208 _x_1 int32
graph.target=14582608_x_1 int32 graph.edge_weight=14582608_x_1 single
initial_classif'=1324310_x_5 single ground_truth=1324310_x_1 unit8
when running function : edgeStruct = UGM_makeEdgeStruct(adjacency,nClasses,1);
adjacency=1324310_x_1324310 nClasses=5,

can you give me some suggestions? can you tell me which step is wrong?
best wishes

[loicland]

Can you repalce
edgeStruct = UGM_makeEdgeStruct(adjacency,nClasses,1)
by
edgeStruct = UGM_makeEdgeStruct(adjacency,nClasses,0)

and add two prints in UGM_makeEdgeStruct:

...
nNodes = int32(length(adj));
fprintf(nNodes )
[i j] = ind2sub([nNodes nNodes],find(adj));
nEdges = length(i)/2;
fprintf(nEdges )
...

[TJ1451154]

hello, when running edgeStruct = UGM_makeEdgeStruct(adjacency,nClasses,0)
'''''
nNodes = int32(length(adj))
%fprintf(nNodes )
[i j] = ind2sub([nNodes nNodes],find(adj));
nEdges = length(i)/2
%fprintf(nEdges)
''''
nNodes =
int32
1324310

nEdges =
7291304

Error using UGM_Infer_LBPC
Requested 21474836485x139689523626408x139689042948512 (17179869184.0GB) array exceeds maximum array size preference. Creation of arrays greater than this limit may take a
long time and cause MATLAB to become unresponsive. See array size limit or preference panel for more information.

Error in UGM_Infer_LBP (line 4)
[nodeBel,edgeBel,logZ] = UGM_Infer_LBPC(nodePot,edgePot,int32(edgeStruct.edgeEnds),int32(edgeStruct.nStates),int32(edgeStruct.V),int32(edgeStruct.E),edgeStruct.maxIter);

Error in LBP_sum_product (line 30)
[nodeBelLBP] = UGM_Infer_LBP(unary,binary,edgeStruct);

Error in benchmark (line 37)
[p_lpb_sp, T] = LBP_sum_product(initial_classif, graph, .5, 10);

[TJ1451154]

Thank you very much for your reply, can you give me some further suggestions?
best wishes

[loicland]

Have you tried repalcing

edgeStruct = UGM_makeEdgeStruct(adjacency,nClasses,1)

by

edgeStruct = UGM_makeEdgeStruct(adjacency,nClasses,0)

I hope that the MALAB error would be more explicit than the C error.

[TJ1451154]

Thank you very much for your help.
First,I run the following code:
graph = build_graph_structure('oakland.ply',10,0,0);
load('oakland_RF', 'initial_classif');
load('oakland_GT', 'ground_truth'); %MUST BE UINT8
[p_lpb_sp, T] = LBP_sum_product(initial_classif, graph, .5, 10);
the values are as follows:

     initial_classif'=1324310_x_5 single 
     ground_truth=1324310_x_1 unit8

Then, step in function LBP_sum_product, and running following code :
'''
nClasses = size(initial_labeling,2);
%it is necessary to have a symetric structure for this algorithm
adjacency = sparse(double(graph.source)+1,double(graph.target)+1 ...
, double(graph.edge_weight));
edgeStruct = UGM_makeEdgeStruct(adjacency,nClasses,0);
unary = initial_labeling;
transition = single(ones(nClasses));
transition(logical(eye(nClasses))) = single(exp(lambda));
binary = repmat(transition, [1 1 edgeStruct.nEdges]);
edgeStruct.useMex = 1;
edgeStruct.maxIter = int32(maxIte);
tic;
clear('V','adj')
[nodeBelLBP] = UGM_Infer_LBP(unary,binary,edgeStruct);
''''
the values are as follows:
in UGM_makeEdgeStruct: adjacency=1324310_x_1324310 nClasses=5,

   in UGM_Infer_LBP:   unary=1324310_x_5 single martix,
                                      binary=5_x_5_x_7291304
                                  edgeStruct in attachment

Finally, when running function UGM_Infer_LBPC
Error using UGM_Infer_LBPC
Requested 21474836485x140149085127080x140149207314848 (17179869184.0GB) array exceeds maximum array size preference. Creation of arrays greater than this limit may take a
long time and cause MATLAB to become unresponsive. See array size limit or preference panel for more information.
Error in UGM_Infer_LBP (line 4)
[nodeBel,edgeBel,logZ] = UGM_Infer_LBPC(nodePot,edgePot,int32(edgeStruct.edgeEnds),int32(edgeStruct.nStates),int32(edgeStruct.V),int32(edgeStruct.E),edgeStruct.maxIter);

Error in LBP_sum_product (line 30)
[nodeBelLBP] = UGM_Infer_LBP(unary,binary,edgeStruct);

Error in benchmark (line 37)
[p_lpb_sp, T] = LBP_sum_product(initial_classif, graph, .5, 10);

[loicland]

Right, so makeEdgeStruct seems to be working correctly.

Can you replace line 26 of LBP_sum_product:

edgeStruct.useMex  = 1;

by

edgeStruct.useMex  = 0;

to again have more informative errors.

Then in GM/infer/UGM_Infer_LBP, could you add just before
new_msg = UGM_LoopyBP(nodePot,edgePot,edgeStruct,maximize);
the following prints:

fprintf(V);
fprintf(E);
fprintf(nEdges);
fprintf(nStates(0:10));

[TJ1451154]

Thank you very much for your help.
I replaced line 26 of LBP_sum_product, edgeStruct.useMex = 0;
when running function UGM_Infer_LBPC,
before function new_msg = UGM_LoopyBP(nodePot,edgePot,edgeStruct,maximize)
the values are as follows:



when running function new_msg = UGM_LoopyBP(nodePot,edgePot,edgeStruct,maximize);
The matlab displays busy all the time, and does not display errors.

[loicland]

Do you end up with a satisfying results? How long oes it take? I'm guessing a long time.

If the matlab versions woks but the C version not, most likely there is a bug. You could try to add:

mexPrintf(nNodes);
mexPrintf(maxState );
mexPrintf(nEdges);

in /UGM/mex/UGM_infer_LBPC.c and recompile the mex files.

[TJ1451154]

1.wait about 20minutes and don"t have any results.
2. I add the code and recompile the mex files.
The matlab still displays busy all the time, and does not display errors.

[loicland]

interupt the matlab code, it is going to take ages. Set the maxIte to 1 and relaunch it.

edgeStruct.maxIter = int32(maxIte);

It would be interesting to see if the MATLAB code function. Also, monitor your RAM use while doing it.

[TJ1451154]

Thank you very much for your help.
when set maxIte to 1, matlab doesn't report errors
but matlab display : Loopy did not converge
Loopy did not converge
CLASSE PRECISION RECALL F_SCORE
1 0.1 100.0 0.1
2 NaN 0.0 NaN
3 NaN 0.0 NaN
4 NaN 0.0 NaN
5 NaN 0.0 NaN
MEAN NaN 20.0 0.1
KAPPA = 0.0 ACCURACY = 0.1
moreover, the cpu usage of matlab is 100%