Evaluated metrics are not meeting expectations

[MR-hyj]

I trained the model for many times using the same metrics as proposed in RPMNet paper. However, I'm not getting an promising evaluating results. Here is one of the evaluated results.

# this model is trained on --noise_type clean --num_points 717

{
    "r_rmse": 30.818792915605933, 
    "r_mae": 23.835809988770535, 
    "t_rmse": 0.3206479847431183, 
    "t_mae": 0.25871723890304565, 
    "err_r_deg_mean": 46.72205352783203, 
    "err_r_deg_rmse": 51.79454803466797, 
    "err_t_mean": 0.5199803709983826, 
    "err_t_rmse": 0.5553786158561707, 
    "chamfer_dist": 0.12541979551315308
}

All metrics are so far away from those in paper:

I don't think I've changed the codes which calculates the metrics. But however I tried, the results were just not good enough.

[yewzijian]

This is weird. The model in Table 1 is trained using 1024 points. Nevertheless, even with 717 points the model should still deliver good accuracies. I just did a quick retraining with 717 points. In the first validation run after 4 training epochs, the mean validation rotation error is already below 2 deg. The test set accuracy is also similar.

You might want to check the code.

[MR-hyj]

This is weird. The model in Table 1 is trained using 1024 points. Nevertheless, even with 717 points the model should still deliver good accuracies. I just did a quick retraining with 717 points. In the first validation run after 4 training epochs, the mean validation rotation error is already below 2 deg. The test set accuracy is also similar.

You might want to check the code.

I'm truly grateful to your efforts and reply. I re-tested the matric-calculating codes with both rot_mag and trans_mag is 0, but still not working. I am checking the codes again. Thank you!