visage_sc: ViSAGe-SC trained on YT-Ambigen (elevation-first ordering, rotation augmentation), epoch 16 -- the model reported in the IJCV paper.
visage_sc_vgg: VGGSound-pretrained initialization used to train the above.
unicodec/final.ckpt: UniCodec fine-tuned for video-to-ambisonics generation, with optimizer and callback state stripped for inference.
