Hi, I have a small dataset (300 training images). The model performs ok during round 1 of training, but when introducing the noise module, the prediction loss returns to where it started in round 1 and does not improve. The noise loss does not decrease over time either. Any ideas? I'm using two sources of noisy ground truth.
Hi, I have a small dataset (300 training images). The model performs ok during round 1 of training, but when introducing the noise module, the prediction loss returns to where it started in round 1 and does not improve. The noise loss does not decrease over time either. Any ideas? I'm using two sources of noisy ground truth.