search

guybenyosef / EchoGraphs

[MICCAI 2022] The official repository of Light-weight spatio-temporal graphs for segmentation and ejection fraction prediction in cardiac ultrasound. Project page: https://guybenyosef.github.io/EchoGraphs/
MIT License
30 stars 8 forks source link

I can't exactly reproduce the results in the paper #7

Open MuchiBai opened 1 year ago

MuchiBai commented 1 year ago

In the case of using the default parameters, the kptsErr of the model is 2.08 +- 1.53, which is significantly different from the 2.3 +- 1.0 in the paper. Could you please provide the detailed hyper-parameters that can exactly reproduce the results in the paper?

SarTho9 commented 1 year ago

Hi, thank you for pointing it out! We reworked the repo to make it more comprehensive. For that purpose we changed the eval kpts error to be the percentage. Because we thought this is better comparable. But in the paper we used the kpts error relative to the pixel size for all of the models. So the comparison in the paper is still correct. If you multiply your results by 1.12 (100 -> 112) you should be able to reproduce the results. This multiplication is done in the kpts_dist_error. We apologize for the inconvinience and will add a remark in the readme to avoid further problems.