Upload 17 files

.gitattributes

@@ -35,3 +35,12 @@ submission/online/submission2/csv_files.tar.gz filter=lfs diff=lfs merge=lfs -te
 submission/online/submission2/raw_results/decoded_labels.pkl filter=lfs diff=lfs merge=lfs -text
 submission/online/submission3/csv_files.tar.gz filter=lfs diff=lfs merge=lfs -text
 submission/online/submission3/raw_results/decoded_labels.pkl filter=lfs diff=lfs merge=lfs -text
+submission/offline/submission1/csv_files.tar.gz filter=lfs diff=lfs merge=lfs -text
+submission/offline/submission1/raw_results/decoded_labels.pkl filter=lfs diff=lfs merge=lfs -text
+submission/offline/submission2/csv_files.tar.gz filter=lfs diff=lfs merge=lfs -text
+submission/offline/submission2/raw_results/decoded_labels.pkl filter=lfs diff=lfs merge=lfs -text
+submission/offline/submission3/csv_files.tar.gz filter=lfs diff=lfs merge=lfs -text
+submission/offline/submission3/raw_results/decoded_labels.pkl filter=lfs diff=lfs merge=lfs -text
+submission/videos/NoXI_019_2016-04-20_Paris_Expert_video-13.mp4 filter=lfs diff=lfs merge=lfs -text
+submission/videos/NoXI_023_2016-04-25_Paris_Expert_video-25.mp4 filter=lfs diff=lfs merge=lfs -text
+submission/videos/RECOLA_group-2_P41-2.mp4 filter=lfs diff=lfs merge=lfs -text

submission/offline/submission1/csv_files.tar.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:220f68b30c6baaea676419090937c7284e3dc1b03fb13bbeabba151e10d3fe5d
+size 118754188

submission/offline/submission1/raw_results/[03-15]-[11-43]-log.log

@@ -0,0 +1,28 @@
+2024-03-15 11:43:41,361 INFO Devices:
+cuda:0
+2024-03-15 11:43:41,361 INFO Devices:
+cuda:4
+2024-03-15 11:43:41,361 INFO Generation config:
+seed: 42
+k: 10
+top_k: 10
+top_p: 0.5
+temperature: 0.67
+generation_task: offline
+with_latent: False
+use_mask_prediction_for_offline_task: False
+load_ckpt: True
+ckpt_dir: results/stage_2/lr8e-05-bs16-schedulerTrue-use_posTrue-mixpbf16-num_devices4/stage2_steps5600-best_val_loss11.8301
+fp16: False
+bf16: True
+phase: test
+generation_batch_size: 8
+generate_results: True
+generate_baselines: False
+
+2024-03-15 11:43:41,361 INFO Devices:
+cuda:3
+2024-03-15 11:43:41,361 INFO Devices:
+cuda:2
+2024-03-15 11:43:41,361 INFO Devices:
+cuda:1

submission/offline/submission1/raw_results/decoded_labels.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a2bc75ac550f7e2528795c1d3230d3179f8d1385dac1f267dcdf717f92b24be0
+size 8027991223

submission/offline/submission1/raw_results/generation_config.yaml

@@ -0,0 +1,72 @@
+dataset_params:
+  ###################### Basic dataset parameters ######################
+  raw_data_dir: /data/lzj/datasets/REACT24_raw  # Change here!
+  base_dir: /data/lzj/projects/REACT24_code  # Change here!
+  cached_data_dir: data
+  checkpoint_dir: checkpoints
+  log_dir: results
+  num_frames: 750
+  feature_extractor_model: checkpoints/preprocessor/feature_extractor
+  fps: 25
+
+
+model_params:
+  max_generation_steps: 750
+  d_model: 1280  # hidden size in encoder and decoder
+  coeff_features_dim: 58  # extracted FaceVerse features dim
+  AU_features_dim: 25088  # extracted AU features dim
+  VA_features_dim: 1408  # extracted VA features dim
+  exp_features_dim: 512  # extracted exp features dim
+  audio_features_dim: 594  # extracted audio features dim
+  max_length: 750  # max length of content
+  vocab_size_3DMM: 80000  # num of quantified 3DMM tokens
+  vocab_size_VA: 2062  # num of quantified VA tokens
+  vocab_size_AU: 4477  # num of quantified AU tokens
+  vocab_size_exp: 8000  # num of quantified expression tokens
+  intermediate_size: 2048  # Intermediate size
+  max_position_embeddings: 1000  # max length of RoPE
+  attention_key_size: 256  # attention intermediate size in GAU
+  num_prior_GAU_layers: 4  # num of GAU layers to fuse 3DMM and prior features
+  num_audio_GAU_layers: 4  # num of GAU layers to fuse 3DMM and audio features
+  num_self_GAU_layers: 4  # num of GAU layers for self-attention
+  attention_dropout: 0.15  # dropout rate of attention scores
+  hidden_dropout: 0.15  # dropout rate of attention output
+  use_bias_GAU: False  # whether to use bias in GAU
+  GAU_activation: 'swish'  # activation function in GAU
+  predict_activation: 'gelu'
+  classifier_dropout: 0.2
+  eps: 1.0e-6
+
+  coeff_dim: 512
+  va_dim: 256
+  au_dim: 256
+  exp_dim: 256
+
+  num_decoder_GAU_layers: 12
+  num_latent: 10
+  tau: 0.67 # The parameter of gumbel softmax
+  bow: True
+  positive_reaction_matching: True
+  negative_reaction_matching: True
+
+generation_params:
+  seed: 42
+
+  k: 10  # number of listener reactions to be generated
+  top_k: 10
+  top_p: 0.5
+  temperature: 0.67
+  generation_task: offline  # Offline / Online
+  with_latent: False
+  use_mask_prediction_for_offline_task: False
+
+  load_ckpt: True
+  ckpt_dir: checkpoints\model\stage_2\lr8e-05-bs16-schedulerTrue-use_posTrue-mixpbf16-num_devices4\stage2_steps5600-best_val_loss11.8301
+
+  fp16: False
+  bf16: True
+  phase: test  # Generate reactions on which dataset, train / val / test
+  generation_batch_size: 8
+
+  generate_results: True
+  generate_baselines: False

submission/offline/submission1/raw_results/log.log

@@ -0,0 +1,10 @@
+2024-03-15 14:53:51,723 INFO Processing the decoded results...
+2024-03-15 15:10:32,770 INFO Computing the metrics...
+2024-03-15 15:10:32,805 INFO start compute FRD
+2024-03-15 15:24:52,717 INFO FRD: 93.96771227392342
+2024-03-15 15:24:52,717 INFO start compute FRC
+2024-03-15 15:28:18,100 INFO FRC: 0.1663991143495104
+2024-03-15 15:28:23,363 INFO start compute TLCC
+2024-03-15 15:28:46,797 INFO TLCC: 44.70905685424805
+2024-03-15 15:28:46,797 INFO Metric: | FRC: 0.16640 | FRD: 93.96771 | FRDiv: 0.10184 | FRVar: 0.03248 | FRDvs: 0.09999 | FRSyn: 44.70906
+2024-03-15 15:28:46,797 INFO Total time: 0:18:13.991987

submission/offline/submission2/csv_files.tar.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc24cf6076e0c8d746833931c3a1e521b6a8a8aa7e20f0375069b9afb9314dcb
+size 352106925

submission/offline/submission2/raw_results/[03-12]-[21-51]-log.log

@@ -0,0 +1,22 @@
+2024-03-12 21:51:28,907 INFO Devices:
+cuda:1
+2024-03-12 21:51:28,918 INFO Devices:
+cuda:0
+2024-03-12 21:51:28,919 INFO Generation config:
+seed: 42
+k: 10
+top_k: 5
+top_p: 0.5
+temperature: 0.2
+generation_task: offline
+with_latent: True
+use_mask_prediction_for_offline_task: False
+load_ckpt: True
+ckpt_dir: results/stage_2/lr8e-05-bs16-schedulerTrue-use_posTrue-mixpbf16-num_devices4/stage2_steps5600-best_val_loss11.8301
+fp16: False
+bf16: True
+phase: test
+generation_batch_size: 4
+generate_results: True
+generate_baselines: False
+

submission/offline/submission2/raw_results/decoded_labels.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ed4b5c6709f4b7b0bd3cafcb152f5eff13144e8153bbdb0184157eae621e5573
+size 8027991223

submission/offline/submission2/raw_results/generation_config.yaml

@@ -0,0 +1,72 @@
+dataset_params:
+  ###################### Basic dataset parameters ######################
+  raw_data_dir: /data/lzj/datasets/REACT24_raw  # Change here!
+  base_dir: /data/lzj/projects/REACT24_code  # Change here!
+  cached_data_dir: data
+  checkpoint_dir: checkpoints
+  log_dir: results
+  num_frames: 750
+  feature_extractor_model: checkpoints/preprocessor/feature_extractor
+  fps: 25
+
+
+model_params:
+  max_generation_steps: 750
+  d_model: 1280  # hidden size in encoder and decoder
+  coeff_features_dim: 58  # extracted FaceVerse features dim
+  AU_features_dim: 25088  # extracted AU features dim
+  VA_features_dim: 1408  # extracted VA features dim
+  exp_features_dim: 512  # extracted exp features dim
+  audio_features_dim: 594  # extracted audio features dim
+  max_length: 750  # max length of content
+  vocab_size_3DMM: 80000  # num of quantified 3DMM tokens
+  vocab_size_VA: 2062  # num of quantified VA tokens
+  vocab_size_AU: 4477  # num of quantified AU tokens
+  vocab_size_exp: 8000  # num of quantified expression tokens
+  intermediate_size: 2048  # Intermediate size
+  max_position_embeddings: 1000  # max length of RoPE
+  attention_key_size: 256  # attention intermediate size in GAU
+  num_prior_GAU_layers: 4  # num of GAU layers to fuse 3DMM and prior features
+  num_audio_GAU_layers: 4  # num of GAU layers to fuse 3DMM and audio features
+  num_self_GAU_layers: 4  # num of GAU layers for self-attention
+  attention_dropout: 0.15  # dropout rate of attention scores
+  hidden_dropout: 0.15  # dropout rate of attention output
+  use_bias_GAU: False  # whether to use bias in GAU
+  GAU_activation: 'swish'  # activation function in GAU
+  predict_activation: 'gelu'
+  classifier_dropout: 0.2
+  eps: 1.0e-6
+
+  coeff_dim: 512
+  va_dim: 256
+  au_dim: 256
+  exp_dim: 256
+
+  num_decoder_GAU_layers: 12
+  num_latent: 10
+  tau: 0.67 # The parameter of gumbel softmax
+  bow: True
+  positive_reaction_matching: True
+  negative_reaction_matching: True
+
+generation_params:
+  seed: 42
+
+  k: 10  # number of listener reactions to be generated
+  top_k: 5
+  top_p: 0.5
+  temperature: 0.2
+  generation_task: offline  # Offline / Online
+  with_latent: True
+  use_mask_prediction_for_offline_task: False
+
+  load_ckpt: True
+  ckpt_dir: checkpoints\model\stage_2_from_scratch\lr8e-05-bs16-schedulerTrue-use_posTrue-mixpbf16-num_devices4\stage2_steps5600-best_val_loss11.8301
+
+  fp16: False
+  bf16: True
+  phase: test  # Generate reactions on which dataset, train / val / test
+  generation_batch_size: 4
+
+  generate_results: True
+  generate_baselines: False

submission/offline/submission2/raw_results/log.log

@@ -0,0 +1,10 @@
+2024-03-15 16:59:52,090 INFO Processing the decoded results...
+2024-03-15 17:09:15,979 INFO Computing the metrics...
+2024-03-15 17:09:16,017 INFO start compute FRD
+2024-03-15 17:20:51,119 INFO FRD: 140.48392646324652
+2024-03-15 17:20:51,119 INFO start compute FRC
+2024-03-15 17:23:42,259 INFO FRC: 0.13901766021162232
+2024-03-15 17:23:46,660 INFO start compute TLCC
+2024-03-15 17:24:09,534 INFO TLCC: 44.1817626953125
+2024-03-15 17:24:09,535 INFO Metric: | FRC: 0.13902 | FRD: 140.48393 | FRDiv: 0.30586 | FRVar: 0.14088 | FRDvs: 0.28812 | FRSyn: 44.18176
+2024-03-15 17:24:09,535 INFO Total time: 0:14:53.518636

submission/offline/submission3/csv_files.tar.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8579efa451253d3f91c897fdd3e4b5698433c5891829c2e2d7d1da69993a4930
+size 150301324

submission/offline/submission3/raw_results/decoded_labels.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5bdf1597ccea8a5895dbeaa5a87d140fc3657acdd5ad5720febe08d482b14aff
+size 8027991223

submission/offline/submission3/raw_results/log.log

@@ -0,0 +1,10 @@
+2024-03-15 18:36:56,817 INFO Processing the decoded results...
+2024-03-15 18:52:57,675 INFO Computing the metrics...
+2024-03-15 18:52:57,727 INFO start compute FRD
+2024-03-15 19:04:31,652 INFO FRD: 104.04324893677952
+2024-03-15 19:04:31,653 INFO start compute FRC
+2024-03-15 19:07:22,701 INFO FRC: 0.1691990281684177
+2024-03-15 19:07:27,914 INFO start compute TLCC
+2024-03-15 19:07:52,382 INFO TLCC: 44.70905685424805
+2024-03-15 19:07:52,383 INFO Metric: | FRC: 0.16920 | FRD: 104.04325 | FRDiv: 0.16754 | FRVar: 0.05353 | FRDvs: 0.13847 | FRSyn: 44.70906
+2024-03-15 19:07:52,383 INFO Total time: 0:14:54.656407

submission/videos/NoXI_019_2016-04-20_Paris_Expert_video-13.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b2220273487af3df0c0ed6bff0c11dd0a027e9fe88d481a6d9ca6fb6e4fc7ba1
+size 1409079

submission/videos/NoXI_023_2016-04-25_Paris_Expert_video-25.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a95d3fb49f20f076a70e2d9c32995d5bc878f29081612c1b79e7ad3884ca6a79
+size 1272319

submission/videos/RECOLA_group-2_P41-2.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8098fc7088803b8daf6c8c11d5c274c91d07584afd12f29fe542baa35b03c9ed
+size 2541453