The reasonable subset gt

[fyture]

Hi，The same result evaluated by your Reasonable-mat is different from the Reasonable-mat that CSP offered. What did I miss? Thanks

[hasanirtiza]

Which model did you evaluate ? How much is the discrepancy ?

[fyture]

Which model did you evaluate ? How much is the discrepancy ?

About Caltech , my own model evaluated by CSP reasonable gt is 13 while using your reasonable gt is 11. And also the size of these two gt files are different. So I don't know what I miss. By the way, where could I get these different subset gt files. I'm a beginner in this field.Appreciate your wonderful work!

[hasanirtiza]

I did not find any difference in sizes for CSP reasonable mat and Pedestron's reasonable mat, both have outer dimensions 1x4024 and also inside each cell, the dimensions are exactly the same. Could you double check, if you are doing everything correctly.

Secondly, you do not need any gt file, you can generate gt files directly from the code by yourself.
Download the new annotations for caltech from here.

Uncomment line 30 in dbEval.m, provide the path variable (pth.annDir=['my folder/anno_test_1xnew'];) to the txt annotations for the testing set, the code will generate the gt file.
Alternatively, have look here as well #5, on how to generate for other settings.

[fyture]

I did not find any difference in sizes for CSP reasonable mat and Pedestron's reasonable mat, both have outer dimensions 1x4024 and also inside each cell, the dimensions and are exactly the same. Could you double check, if you are doing everything correctly.

Secondly, you do not need any gt file, you can generate gt files directly from the code by yourself.
Download the new annotations for caltech from here.

Uncomment line 30 in dbEval.m, provide the path(variable pth.annDir=['my folder/anno_test_1xnew'];) to the txt annotations for the testing set, the code will generate the gt file.
Alternatively, have look here as well #5, on how to generate for other settings.

Thank you so much!!!!

[fyture]

I did not find any difference in sizes for CSP reasonable mat and Pedestron's reasonable mat, both have outer dimensions 1x4024 and also inside each cell, the dimensions and are exactly the same. Could you double check, if you are doing everything correctly.
Secondly, you do not need any gt file, you can generate gt files directly from the code by yourself.
Download the new annotations for caltech from here.
Uncomment line 30 in dbEval.m, provide the path(variable pth.annDir=['my folder/anno_test_1xnew'];) to the txt annotations for the testing set, the code will generate the gt file.
Alternatively, have look here as well #5, on how to generate for other settings.

Thank you so much!!!!

Really sorry for bothering you again. I followed your guide and successfully got the reasults of Reasonable, Heavy and All subsets. But other than these three subsets, The reasults of other subsets, say Scale=large, is NAN. What went wrong? Don‘t the new annotations support other subets or some other reasons? Thanks again!!!!!!!!!!

[hasanirtiza]

Hard to comment that why exactly are you getting NaN. First check if there are any samples in the ground-truth file belonging to Scale=Large. Play with the visibility parameter, which is by default set to [inf inf], may be start by setting it to [0.65 inf].