Error in SAD_blockmatching

nmajid2 commented 3 years ago

Hello, I've just tried the code and it works with no error based on the default settings. However, when I changed the code in Edgeflow_prepare_data.m and replace the 'forward_camera' with 'downward_camera' and got the following error:

Index exceeds the number of array elements (128).

Error in SAD_blockmatching (line 35)
        SAD_temp(r+D+1)=sum(abs(hist_previous(x-W+pixel_shift:x+W+pixel_shift)-hist_current(x+r-W:x+r+W)));

Error in edge_flow_v2 (line 50)
[displacement.x, match_error.x,
fit_quality.x]=SAD_blockmatching(window,max_search_distance,edge_histogram(1).x,
edge_histogram(frame_previous_number.x+1).x,derot.x,1);

Error in calcEdgeFlow (line 33)
    edge_flow_v2(Fx_hist_left,Fy_hist_left,edge_histogram,previous_parameters,...

Error in calculate_OF_on_images (line 72)
        calcEdgeFlow

Error in Edgeflow_generate_results (line 61)
        calculate_OF_on_images

I try to fix the issue, but in case you have the solution in mind I appreciate your help. Thank you very much.

knmcguire commented 3 years ago

Hi!

It will be difficult for me to test this out for you since I no longer have access to matlab. Probably I should also upload some python version of this if I will find time, since it is difficult to debug this without the proper tools...

Have you checked the size of the downward camera's images? maybe they are not 128 px long?

nmajid2 commented 3 years ago

Hi,

I realized the error also happens for forward_camera. The way to reproduce the error is to change chosen_tracks(2).track = [2] to chosen_tracks(2).track = [7] in the edgeflow_generate_results.m.

After debugging more of the code, it sounds that the problem is here:

SAD_temp(r+D+1)=sum(abs(hist_previous(x-W+pixel_shift:x+W+pixel_shift)-hist_current(x+r-W:x+r+W)));

Sometimes the value of x+W+pixel_shift goes above 131, some other times the values of x-W+pixel_shift become -12 and maybe less. So the root of the issue might be the calculation of pixel_shift, because its values based on my observation become a negative number like -28. The boundaries for hist_previous and hist_current are [1,128].

I checked the size of the images, they all have the same size, 256 X 96 pixels.

knmcguire commented 3 years ago

hmmm that could be but I haven't see this happen before, and it should be limiters in the code to handle this.

I will try to get my hands on a PC with matlab again but this will take some time. In the mean time you can double check those limits that are on there and maybe arrange something for that. However considering my inability to help out I might consider to rewrite everything in python instead in the future.

nmajid2 commented 3 years ago

Thanks a lot, having the code in python would be very helpful. I can also check later to see whether I can use the Matlab Coder to convert the code into c/c++. And then python binding technologies can be used to access the library from python. For now, as you suggested, I keep debugging the Matlab code, since fixing the error in Matlab is easier than in c/c++.

knmcguire / EdgeFlow_matlab

Error in SAD_blockmatching #6