Add ViSAGe-SC (IJCV) checkpoints and fine-tuned UniCodec

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.

unicodec/final.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5668b926f6a3c8e7ca73f5ca3179a2c5fa4df57d2f9bde207e5f69c7367ed01e
+size 1351219409

visage_sc/config.json

@@ -0,0 +1,58 @@
+{
+  "_name_or_path": "/gallery_millet/jaeyeon.kim/ijcv_results/unicodec_checkpoint/vgg_final",
+  "activation_dropout": 0.0,
+  "activation_function": "gelu",
+  "attention_dropout": 0.0,
+  "bos_token_id": 0,
+  "cfg_dropout_prob": 0.1,
+  "cfg_guidance_scale": 2.5,
+  "cfg_spatial_guidance_scale": 2.5,
+  "classifier_dropout": 0.0,
+  "classifier_free_guidance": true,
+  "clip_frame_rate": 4,
+  "d_clip": 512,
+  "d_model": 1024,
+  "dac_downsample_rate": 320,
+  "dac_frame_rate": 75,
+  "dac_pad_token_id": 16384,
+  "dac_sample_rate": 24000,
+  "dac_vocab_size": 16384,
+  "decoder_attention_heads": 16,
+  "decoder_ffn_dim": 4096,
+  "decoder_layerdrop": 0.0,
+  "decoder_layers": 12,
+  "decoder_start_token_id": 2,
+  "dropout": 0.1,
+  "encoder_attention_heads": 16,
+  "encoder_ffn_dim": 4096,
+  "encoder_layerdrop": 0.0,
+  "encoder_layers": 6,
+  "eos_token_id": 2,
+  "forced_eos_token_id": 2,
+  "id2label": {
+    "0": "LABEL_0",
+    "1": "LABEL_1",
+    "2": "LABEL_2"
+  },
+  "init_additional_from_clip": false,
+  "init_std": 0.02,
+  "is_encoder_decoder": true,
+  "label2id": {
+    "LABEL_0": 0,
+    "LABEL_1": 1,
+    "LABEL_2": 2
+  },
+  "label_smoothing": 0.0,
+  "max_position_embeddings": 1024,
+  "max_source_length": 80,
+  "max_target_length": 400,
+  "model_type": "bart",
+  "num_hidden_layers": 6,
+  "num_rvq": 1,
+  "pad_token_id": 1,
+  "scale_embedding": false,
+  "tie_word_embeddings": false,
+  "transformers_version": "4.40.2",
+  "use_cache": true,
+  "vocab_size": 50265
+}

visage_sc/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2cb353bb6b910b9161241358e1e44049c0eca6b90eee2c2744a59ed18252e097
+size 1659624050

visage_sc_vgg/config.json

@@ -0,0 +1,57 @@
+{
+  "activation_dropout": 0.0,
+  "activation_function": "gelu",
+  "attention_dropout": 0.0,
+  "bos_token_id": 0,
+  "cfg_dropout_prob": 0.1,
+  "cfg_guidance_scale": 2.5,
+  "cfg_spatial_guidance_scale": 2.5,
+  "classifier_dropout": 0.0,
+  "classifier_free_guidance": true,
+  "clip_frame_rate": 4,
+  "d_clip": 512,
+  "d_model": 1024,
+  "dac_downsample_rate": 320,
+  "dac_frame_rate": 75,
+  "dac_pad_token_id": 16384,
+  "dac_sample_rate": 24000,
+  "dac_vocab_size": 16384,
+  "decoder_attention_heads": 16,
+  "decoder_ffn_dim": 4096,
+  "decoder_layerdrop": 0.0,
+  "decoder_layers": 12,
+  "decoder_start_token_id": 2,
+  "dropout": 0.1,
+  "encoder_attention_heads": 16,
+  "encoder_ffn_dim": 4096,
+  "encoder_layerdrop": 0.0,
+  "encoder_layers": 6,
+  "eos_token_id": 2,
+  "forced_eos_token_id": 2,
+  "id2label": {
+    "0": "LABEL_0",
+    "1": "LABEL_1",
+    "2": "LABEL_2"
+  },
+  "init_additional_from_clip": false,
+  "init_std": 0.02,
+  "is_encoder_decoder": true,
+  "label2id": {
+    "LABEL_0": 0,
+    "LABEL_1": 1,
+    "LABEL_2": 2
+  },
+  "label_smoothing": 0.0,
+  "max_position_embeddings": 1024,
+  "max_source_length": 80,
+  "max_target_length": 400,
+  "model_type": "bart",
+  "num_hidden_layers": 6,
+  "num_rvq": 1,
+  "pad_token_id": 1,
+  "scale_embedding": false,
+  "tie_word_embeddings": false,
+  "transformers_version": "4.40.2",
+  "use_cache": true,
+  "vocab_size": 50265
+}

visage_sc_vgg/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1fa9c5ae803ad5eb0903919ee0db03d64be65cc913d627de37ddfd6649115123
+size 1247362662