Trained Model

Weights of all the encoders

1. streamline encoder (CNN) frozen
2. patch encoder (dVAE) frozen
3. cluster encoder (pointnet) learnable

Where tractoembed_v1.pth is equivalent to cluster encoder in terms of weights till MECL. And tractoembed_v1.pth contains final FC layers weights also, along with cluster encoder weights.

As mentioned, streamline and patch encoders are trained separately