Sample image of pinna contours

[babar414]

Hi, Thank you for sharing your code.

I was wondering if it would be possible for you to share a sample image of pinna contours so I know that I can prepare my input image accordingly. (Perhaps one of the image from training or validation dataset would be good)

Thanks

[AudioEngineer99]

Hi,

I would also be very interested in that.

I generated several contour images with different settings of MATLAB's 'edge' function, leading to more or less detailed contours. So far, the results have been very inaccurate, i.e. high mismatch of ILD in the left halfplane and "non-symmetric" HRTF magnitudes for left and right ear.

Thanks.

[AudioEngineer99]

So, I tested the model on subject 2 of the HUTUBS database [1]. I used three different images of the ear:

Using, head width, head depth, and shoulder circumference of subject 2, as given in the HUTUBS database, I get the following HRTF magnitudes in the horizontal plane:

, where the blue curve corresponds to the left image, the orange one to the center image, and the yellow curve to the right grayscale image.

Given that there are relatively high differences in the images, I find it very odd that the model produces almost identical magnitudes.

If I understood correctly, the model was trained on the HUTUBS database. Therefore, I would expect that it should be easy to reproduce the HRTFs of the training set. Maybe something went wrong in the implementation due to changes of Python packages (as disscussed here)?

[1] F. Brinkmann, M. Dinakaran, R. Pelzer, P. Grosche, D. Voss, and S. Weinzierl, “A cross-evaluated database of measured and simulated HRTFs including 3d head meshes, anthropometric features, and headphone impulse responses,” AES J. Audio Eng. Soc., vol. 67, no. 9, pp. 705–718, 2019, doi: 10.17743/jaes.2019.0024.

[pandagre]

Hi,
I also encountered inaccurate results, specifically with high mismatch of Interaural Level Difference (ILD) in the left half-plane, and non-symmetric Head-Related Transfer Function (HRTF) magnitudes for the left and right ears.
Has anyone encountered similar issues with mismatched ILDs in the left half-plane, or non-symmetric HRTF magnitudes?
What are some common causes or solutions for this kind of behavior, particularly to the model’s handling of ILD and symmetry?
Please help!

Thanks