PointNet is not being used

[mpegia]

Hello and thank you for sharing your code! You did a great work.

I noticed a couple of issues in the code. I would greatly appreciate your help in resolving them.

1. The code provided did not use the PointNet network (DGCNN) in the forward function of the CorrNet. Could you please update the code or include the missing patch to the CorrNet network.
2. When training the model using only the image and mesh modalities for 10 epochs, you achieved a high mean average precision (mAP) of 70.96. However, the precision, and F-score calculated for the same run were very low while the recall too high (precision=0.035, recall=0.9985, f-score=0.067). This suggests that while the model performs well in ranking relevant results higher, it may also retrieve a significant number of false positives.

Best regards, Maria