The uploaded model is of poor quality. It doesn't look like a MAP at 5 = 0.9+
#1
by
alex8mlhf
- opened
I've checked over 10 images on your inference. And in half of the images, the dots are either very bad or lie within 10-20 pixels of the desired accuracy.
Hi Alex,
Thanks for your feedback. You are correct but we need to take into account that the model has been trained on a very small number of frames, so that difference in pixel position is something understandable.
We would really appreciate it if you are interested in this project to contribute on expanding the dataset and make the model more generalizable across different settings.
MehdiH7
changed discussion status to
closed