chore: remove non-weight files (batch 6)

video/dfd-fcg/model_code/configs/robustness/CC(3).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/CC/3/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/CC(4).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/CC/4/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/CC(5).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/CC/5/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/CS(1).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/CS/1/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/CS(2).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/CS/2/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/CS(3).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/CS/3/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/CS(4).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/CS/4/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/CS(5).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/CS/5/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/GB(1).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/GB/1/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/GB(2).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/GB/2/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/GB(3).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/GB/3/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/GB(4).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/GB/4/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/GB(5).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/GB/5/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/GNC(1).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/GNC/1/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/GNC(2).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/GNC/2/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/GNC(3).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/GNC/3/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/GNC(4).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/GNC/4/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/GNC(5).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/GNC/5/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/JPEG(1).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/JPEG/1/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/JPEG(2).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/JPEG/2/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/JPEG(3).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/JPEG/3/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/JPEG(4).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/JPEG/4/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/JPEG(5).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/JPEG/5/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/VC(1).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/VC/1/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/VC(2).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/VC/2/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/VC(3).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/VC/3/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/VC(4).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/VC/4/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/robustness/VC(5).yaml

@@ -1,14 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    test_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        force_random_speed: null
-        data_dir: 'datasets/robustness/VC/5/'
-        vid_ext: .avi

video/dfd-fcg/model_code/configs/scenario/LOO/DF.yaml

@@ -1,32 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    train_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        batch_size: 30
-        df_types: ['REAL','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: FORCE_PAIR
-        augmentations:
-        - NORMAL
-        - VIDEO
-        - VIDEO_RRC
-        - FRAME
-        force_random_speed: null
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 3
-    val_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','NT','FS','F2F']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1

video/dfd-fcg/model_code/configs/scenario/LOO/F2F.yaml

@@ -1,32 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    train_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        batch_size: 30
-        df_types: ['REAL','DF','FS','NT']
-        compressions: ['c23']
-        strategy: FORCE_PAIR
-        augmentations:
-        - NORMAL
-        - VIDEO
-        - VIDEO_RRC
-        - FRAME
-        force_random_speed: null
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 3
-    val_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1

video/dfd-fcg/model_code/configs/scenario/LOO/FS.yaml

@@ -1,32 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    train_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        batch_size: 30
-        df_types: ['REAL','DF','F2F','NT']
-        compressions: ['c23']
-        strategy: FORCE_PAIR
-        augmentations:
-        - NORMAL
-        - VIDEO
-        - VIDEO_RRC
-        - FRAME
-        force_random_speed: null
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 3
-    val_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1

video/dfd-fcg/model_code/configs/scenario/LOO/NT.yaml

@@ -1,32 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    train_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        batch_size: 30
-        df_types: ['REAL','DF','FS','F2F']
-        compressions: ['c23']
-        strategy: FORCE_PAIR
-        augmentations:
-        - NORMAL
-        - VIDEO
-        - VIDEO_RRC
-        - FRAME
-        force_random_speed: null
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 3
-    val_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1

video/dfd-fcg/model_code/configs/scenario/compression/c40.yaml

@@ -1,55 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    batch_size: 20
-    num_workers: 4
-    clip_duration: 3
-    num_frames: 10
-    train_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        batch_size: 30
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c40']
-        strategy: FORCE_PAIR
-        augmentations:
-        - NORMAL
-        - VIDEO
-        - VIDEO_RRC
-        - FRAME
-        force_random_speed: null
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 3
-    val_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c40']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1
-    - class_path: src.dataset.cdf.CDFDataModule
-      init_args:
-        data_dir: 'datasets/cdf/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1
-    - class_path: src.dataset.dfdc.DFDCDataModule
-      init_args:
-        data_dir: 'datasets/dfdc/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1
-    - class_path: src.dataset.fsh.FShDataModule
-      init_args:
-        compressions: ['c40']
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1

video/dfd-fcg/model_code/configs/scenario/compression/raw.yaml

@@ -1,55 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    batch_size: 20
-    num_workers: 4
-    clip_duration: 3
-    num_frames: 10
-    train_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        batch_size: 30
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['raw']
-        strategy: FORCE_PAIR
-        augmentations:
-        - NORMAL
-        - VIDEO
-        - VIDEO_RRC
-        - FRAME
-        force_random_speed: null
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 3
-    val_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['raw']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1
-    - class_path: src.dataset.cdf.CDFDataModule
-      init_args:
-        data_dir: 'datasets/cdf/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1
-    - class_path: src.dataset.dfdc.DFDCDataModule
-      init_args:
-        data_dir: 'datasets/dfdc/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1
-    - class_path: src.dataset.fsh.FShDataModule
-      init_args:
-        compressions: ['raw']
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1

video/dfd-fcg/model_code/configs/scenario/partial/10.yaml

@@ -1,56 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    batch_size: 20
-    num_workers: 4
-    clip_duration: 3
-    num_frames: 10
-    train_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        batch_size: 30
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: FORCE_PAIR
-        augmentations:
-        - NORMAL
-        - VIDEO
-        - VIDEO_RRC
-        - FRAME
-        force_random_speed: null
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        ratio: 0.1
-        max_clips: 3
-    val_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1
-    - class_path: src.dataset.cdf.CDFDataModule
-      init_args:
-        data_dir: 'datasets/cdf/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1
-    - class_path: src.dataset.dfdc.DFDCDataModule
-      init_args:
-        data_dir: 'datasets/dfdc/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1
-    - class_path: src.dataset.fsh.FShDataModule
-      init_args:
-        compressions: ['c23']
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1

video/dfd-fcg/model_code/configs/scenario/partial/25.yaml

@@ -1,56 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    batch_size: 20
-    num_workers: 4
-    clip_duration: 3
-    num_frames: 10
-    train_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        batch_size: 30
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: FORCE_PAIR
-        augmentations:
-        - NORMAL
-        - VIDEO
-        - VIDEO_RRC
-        - FRAME
-        force_random_speed: null
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        ratio: 0.25
-        max_clips: 3
-    val_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1
-    - class_path: src.dataset.cdf.CDFDataModule
-      init_args:
-        data_dir: 'datasets/cdf/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1
-    - class_path: src.dataset.dfdc.DFDCDataModule
-      init_args:
-        data_dir: 'datasets/dfdc/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1
-    - class_path: src.dataset.fsh.FShDataModule
-      init_args:
-        compressions: ['c23']
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1

video/dfd-fcg/model_code/configs/scenario/partial/50.yaml

@@ -1,56 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    batch_size: 20
-    num_workers: 4
-    clip_duration: 3
-    num_frames: 10
-    train_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        batch_size: 30
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: FORCE_PAIR
-        augmentations:
-        - NORMAL
-        - VIDEO
-        - VIDEO_RRC
-        - FRAME
-        force_random_speed: null
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        ratio: 0.50
-        max_clips: 3
-    val_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1
-    - class_path: src.dataset.cdf.CDFDataModule
-      init_args:
-        data_dir: 'datasets/cdf/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1
-    - class_path: src.dataset.dfdc.DFDCDataModule
-      init_args:
-        data_dir: 'datasets/dfdc/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1
-    - class_path: src.dataset.fsh.FShDataModule
-      init_args:
-        compressions: ['c23']
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1

video/dfd-fcg/model_code/configs/scenario/partial/75.yaml

@@ -1,56 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    batch_size: 20
-    num_workers: 4
-    clip_duration: 3
-    num_frames: 10
-    train_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        batch_size: 30
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: FORCE_PAIR
-        augmentations:
-        - NORMAL
-        - VIDEO
-        - VIDEO_RRC
-        - FRAME
-        force_random_speed: null
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        ratio: 0.75
-        max_clips: 3
-    val_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1
-    - class_path: src.dataset.cdf.CDFDataModule
-      init_args:
-        data_dir: 'datasets/cdf/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1
-    - class_path: src.dataset.dfdc.DFDCDataModule
-      init_args:
-        data_dir: 'datasets/dfdc/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1
-    - class_path: src.dataset.fsh.FShDataModule
-      init_args:
-        compressions: ['c23']
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1

video/dfd-fcg/model_code/configs/scenario/robust/robust.yaml

@@ -1,33 +0,0 @@
-data:
-  class_path: src.dataset.base.ODDeepFakeDataModule
-  init_args:
-    batch_size: 20
-    num_workers: 4
-    clip_duration: 3
-    num_frames: 10
-    train_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        batch_size: 30
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: FORCE_PAIR
-        augmentations:
-        - ROBUSTNESS
-        force_random_speed: null
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 3
-    val_datamodules:
-    - class_path: src.dataset.ffpp.FFPPDataModule
-      init_args:
-        df_types: ['REAL','DF','FS','F2F','NT']
-        compressions: ['c23']
-        strategy: NORMAL
-        augmentations: 
-        - NONE
-        data_dir: 'datasets/ffpp/'
-        vid_ext: '.avi'
-        pack: false
-        max_clips: 1

video/dfd-fcg/model_code/configs/test.yaml

@@ -1,14 +0,0 @@
-trainer:
-  logger:
-    init_args:
-      offline: true
-  limit_train_batches: 30
-  limit_val_batches: 30
-  accumulate_grad_batches: 1
-data:
-  init_args:
-    batch_size: 1
-    num_workers: 0
-    train_datamodules:
-      - init_args:
-          batch_size: 1

video/dfd-fcg/model_code/demo.py

@@ -1,192 +0,0 @@
-import os
-import cv2
-import sys
-import yaml
-import json
-import math
-import torch
-import pickle
-import shutil
-import logging
-import warnings
-import argparse
-import numpy as np
-
-
-from os import path
-from datetime import datetime
-from torchvision.io import VideoReader
-from src.utility.builtin import ODTrainer, ODLightningCLI
-
-
-def parse_args(args=None):
-    parser = argparse.ArgumentParser()
-    parser.add_argument("model_cfg_path", type=str)
-    parser.add_argument("model_ckpt_path", type=str)
-    parser.add_argument("video_path", type=str)
-    parser.add_argument("--out_path", type=str, default=None)
-    parser.add_argument("--threshold", type=float, default=0.5)
-    parser.add_argument("--precision", type=str, default="16")
-    parser.add_argument("--batch_size", type=int, default=30)
-    return parser.parse_args(args=args)
-
-
-def configure_logging():
-    logging_fmt = "[%(levelname)s][%(filename)s:%(lineno)d]: %(message)s"
-    logging.basicConfig(level="INFO", format=logging_fmt)
-    warnings.filterwarnings(action="ignore")
-
-
-@torch.inference_mode()
-def demo_driver(cli, ckpt_path, video_path, out_path, batch_size, threshold):
-    # setup model
-    model = cli.model
-
-    try:
-        model = model.__class__.load_from_checkpoint(ckpt_path)
-    except Exception as e:
-        print(f"Unable to load model from checkpoint in strict mode: {e}")
-        print(f"Loading model from checkpoint in non-strict mode.")
-        model = model.__class__.load_from_checkpoint(ckpt_path, strict=False)
-
-    model.eval()
-    transforms = model.transform
-
-    BATCH = batch_size
-    stride = 0.333
-
-    # load original video
-    vid_reader = VideoReader(video_path, "video", num_threads=1)
-    vid_ext = os.path.splitext(video_path)[-1]
-    vid_name = os.path.split(video_path)[1].replace(vid_ext, "")
-    fps = vid_reader.get_metadata()["video"]["fps"][0]
-
-    frames = []
-    for frame_data in vid_reader:
-        frames.append(frame_data["data"])
-    frames = torch.stack(frames)
-    del vid_reader
-    _, H, W = frames[0].shape
-
-    # load bboxes of original video
-    with open(video_path.replace("videos", "frame_data").replace(vid_ext, ".pickle"), "rb") as f:
-        fdata = pickle.load(f)
-        bboxes = []
-        for data in fdata:
-            data["bboxes"] = [
-                bbox.reshape(2, -1)
-                if len(bbox.shape) == 1 else bbox
-                for bbox in data["bboxes"]
-            ]
-            face_idx = np.argsort([
-                np.linalg.norm((bbox[0] - bbox[1])) for bbox in data["bboxes"]
-            ])[-1]
-            bboxes.append(data["bboxes"][face_idx])
-
-    # load face cropped video
-    vid_reader = VideoReader(
-        video_path.replace("/videos", "/cropped/videos").replace(vid_ext, ".avi"),
-        "video",
-        num_threads=1
-    )
-    cropped_frames = []
-    for frame_data in vid_reader:
-        cropped_frames.append(frame_data["data"])
-    cropped_frames = torch.stack(cropped_frames)
-    del vid_reader
-
-    # sample frames and inference
-    indices = torch.tensor([int(math.floor(i * stride * fps)) for i in range(10)], dtype=torch.long)
-    probs = []
-    i = 0
-    clip_count = len(cropped_frames) - indices[-1]
-    while (i < clip_count):
-        batch = min(clip_count - i, BATCH)
-        clips = torch.stack([
-            transforms(cropped_frames[indices + i + j]) for j in range(batch)
-        ]).to("cuda")
-        results = model.evaluate(clips)
-        probs.extend(results["logits"].softmax(dim=-1)[:, 1].flatten().cpu().tolist())
-        i += batch
-
-    # draw and write to video
-    bbox_frames = []
-    for frame, bbox, prob in zip(frames[indices[-1]:], bboxes[indices[-1]:], probs):
-        frame = frame.permute(1, 2, 0).numpy()
-        frame = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)
-        thickness = int(np.linalg.norm(bbox[0] - bbox[1]) * 0.01)
-        color = (0, 255, 0) if prob < threshold else (0, 0, 255)
-        category = "REAL" if prob < threshold else "FAKE"
-        frame = cv2.rectangle(
-            frame,
-            bbox[0].astype(int),
-            bbox[1].astype(int),
-            color,
-            thickness
-        )
-        frame = cv2.putText(
-            frame,
-            f'{round(prob,2)}',
-            [int(bbox[0][0]), int(bbox[1][1] - thickness)],
-            cv2.FONT_HERSHEY_SIMPLEX,
-            1, color, thickness, cv2.LINE_AA
-        )
-
-        frame = cv2.putText(
-            frame,
-            category,
-            [int(bbox[0][0]), int(bbox[0][1] - thickness)],
-            cv2.FONT_HERSHEY_SIMPLEX,
-            1, color, thickness, cv2.LINE_AA
-        )
-
-        bbox_frames.append(frame)
-
-    out_path = (f'pred_{vid_name}.avi' if out_path is None else out_path)
-
-    writer = cv2.VideoWriter(
-        out_path,
-        cv2.VideoWriter_fourcc('X', 'V', 'I', 'D'),
-        fps,
-        (W, H)
-    )
-
-    for frame in bbox_frames:
-        writer.write(frame)
-
-    writer.release()
-
-
-if __name__ == "__main__":
-    configure_logging()
-
-    params = parse_args()
-
-    cli = ODLightningCLI(
-        run=False,
-        trainer_class=ODTrainer,
-        save_config_callback=None,
-        parser_kwargs={
-            "parser_mode": "omegaconf"
-        },
-        auto_configure_optimizers=False,
-        seed_everything_default=1019,
-        args=[
-            '-c', params.model_cfg_path,
-            '--trainer.logger=null',
-            f'--trainer.devices=1',
-            f'--trainer.precision={params.precision}',
-        ],
-    )
-
-    ckpt_path = params.model_ckpt_path
-    video_path = params.video_path
-
-    demo_driver(
-        cli=cli,
-        ckpt_path=ckpt_path,
-        video_path=video_path,
-        batch_size=params.batch_size,
-        threshold=params.threshold,
-        out_path=params.out_path
-    )

video/dfd-fcg/model_code/environment.yml

@@ -1,129 +0,0 @@
-name: dfd-fcg
-channels:
-  - defaults
-dependencies:
-  - _libgcc_mutex=0.1
-  - _openmp_mutex=5.1
-  - ca-certificates=2023.12.12
-  - ld_impl_linux-64=2.38
-  - libffi=3.4.4
-  - libgcc-ng=11.2.0
-  - libgomp=11.2.0
-  - libstdcxx-ng=11.2.0
-  - ncurses=6.4
-  - openssl=3.0.13
-  - pip=23.3.1
-  - python=3.8.18
-  - readline=8.2
-  - setuptools=68.2.2
-  - sqlite=3.41.2
-  - tk=8.6.12
-  - wheel=0.41.2
-  - xz=5.4.6
-  - zlib=1.2.13
-  - pip:
-      - aiohttp==3.9.3
-      - aiosignal==1.3.1
-      - albumentations==1.4.0
-      - antlr4-python3-runtime==4.9.3
-      - appdirs==1.4.4
-      - async-timeout==4.0.3
-      - attrs==23.2.0
-      - av==11.0.0
-      - bitsandbytes==0.41.0
-      - certifi==2024.2.2
-      - charset-normalizer==3.3.2
-      - click==8.1.7
-      - contourpy==1.1.1
-      - cycler==0.12.1
-      - docker-pycreds==0.4.0
-      - docstring-parser==0.15
-      - face-alignment==1.4.1
-      - filelock==3.13.1
-      - fonttools==4.49.0
-      - frozenlist==1.4.1
-      - fsspec==2024.2.0
-      - ftfy==6.1.3
-      - gitdb==4.0.11
-      - gitpython==3.1.42
-      - huggingface-hub==0.21.3
-      - hydra-core==1.3.2
-      - idna==3.6
-      - imageio==2.34.0
-      - importlib-resources==6.1.2
-      - jinja2==3.1.3
-      - joblib==1.3.2
-      - jsonargparse==4.27.5
-      - kiwisolver==1.4.5
-      - lazy-loader==0.3
-      - lightning==2.2.0.post0
-      - lightning-utilities==0.10.1
-      - markdown-it-py==3.0.0
-      - markupsafe==2.1.5
-      - matplotlib==3.7.5
-      - mdurl==0.1.2
-      - mpmath==1.3.0
-      - multidict==6.0.5
-      - networkx==3.1
-      - numpy==1.24.4
-      - nvidia-cublas-cu12==12.1.3.1
-      - nvidia-cuda-cupti-cu12==12.1.105
-      - nvidia-cuda-nvrtc-cu12==12.1.105
-      - nvidia-cuda-runtime-cu12==12.1.105
-      - nvidia-cudnn-cu12==8.9.2.26
-      - nvidia-cufft-cu12==11.0.2.54
-      - nvidia-curand-cu12==10.3.2.106
-      - nvidia-cusolver-cu12==11.4.5.107
-      - nvidia-cusparse-cu12==12.1.0.106
-      - nvidia-nccl-cu12==2.19.3
-      - nvidia-nvjitlink-cu12==12.3.101
-      - nvidia-nvtx-cu12==12.1.105
-      - omegaconf==2.3.0
-      - open-clip-torch==2.24.0
-      - opencv-python==4.9.0.80
-      - opencv-python-headless==4.9.0.80
-      - packaging==23.2
-      - pandas==2.0.3
-      - pillow==10.2.0
-      - protobuf==4.25.3
-      - psutil==5.9.8
-      - pygments==2.17.2
-      - pyparsing==3.1.1
-      - python-dateutil==2.9.0.post0
-      - pytorch-lightning==2.2.0.post0
-      - pytz==2024.1
-      - pywavelets==1.4.1
-      - pyyaml==6.0.1
-      - qudida==0.0.4
-      - regex==2023.12.25
-      - requests==2.31.0
-      - rich==13.7.1
-      - safetensors==0.4.2
-      - scikit-image==0.21.0
-      - scikit-learn==1.3.2
-      - scipy==1.10.1
-      - sentencepiece==0.2.0
-      - sentry-sdk==1.40.6
-      - setproctitle==1.3.3
-      - six==1.16.0
-      - smmap==5.0.1
-      - sympy==1.12
-      - tensorboardx==2.6.2.2
-      - threadpoolctl==3.3.0
-      - tifffile==2023.7.10
-      - timm==0.9.16
-      - torch==2.2.1
-      - torchaudio==2.2.1
-      - torchmetrics==1.3.1
-      - torchvision==0.17.1
-      - tqdm==4.66.2
-      - triton==2.2.0
-      - typeshed-client==2.5.1
-      - typing-extensions==4.10.0
-      - tzdata==2024.1
-      - urllib3==2.2.1
-      - wandb==0.16.3
-      - wcwidth==0.2.13
-      - yarl==1.9.4
-      - zipp==3.17.0
-prefix: /home/od/miniconda3/envs/dfd-fcg

video/dfd-fcg/model_code/inference.py

@@ -1,187 +0,0 @@
-import os
-import sys
-import yaml
-import json
-import torch
-import pickle
-import shutil
-import logging
-import warnings
-import argparse
-
-
-from os import path
-from datetime import datetime
-from torchmetrics.classification import AUROC, Accuracy
-from src.utility.builtin import ODTrainer, ODLightningCLI
-
-
-def parse_args(args=None):
-    parser = argparse.ArgumentParser()
-    parser.add_argument("model_cfg_path", type=str)
-    parser.add_argument("data_cfg_path", type=str)
-    parser.add_argument("model_ckpt_path", type=str)
-    parser.add_argument("--precision", type=str, default="16")
-    parser.add_argument("--devices", type=int, default=-1)
-    parser.add_argument("--notes", type=str, default='')
-    return parser.parse_args(args=args)
-
-
-class StatsRecorder:
-    def __init__(self, label):
-        self.label = label
-        self.prob = 0
-        self.count = 0
-
-    def update(self, prob, label):
-        assert label == self.label
-        self.prob += prob
-        self.count += 1
-
-    def compute(self):
-        return {
-            "label": self.label,
-            "prob": self.prob / self.count
-        }
-
-
-def configure_logging():
-    logging_fmt = "[%(levelname)s][%(filename)s:%(lineno)d]: %(message)s"
-    logging.basicConfig(level="INFO", format=logging_fmt)
-    warnings.filterwarnings(action="ignore")
-
-
-@torch.inference_mode()
-def inference_driver(cli, cfg_dir, ckpt_path, notes=None):
-
-    timestamp = datetime.now().strftime("%m%dT%H%M%S")
-
-    trainer = cli.trainer
-
-    # setup model
-    model = cli.model
-
-    try:
-        model = model.__class__.load_from_checkpoint(ckpt_path)
-    except Exception as e:
-        print(f"Unable to load model from checkpoint in strict mode: {e}")
-        print(f"Loading model from checkpoint in non-strict mode.")
-        model = model.__class__.load_from_checkpoint(ckpt_path, strict=False)
-
-    model.eval()
-
-    # setup dataset
-    datamodule = cli.datamodule
-    datamodule.prepare_data()
-    datamodule.affine_model(cli.model)
-    datamodule.setup('test')
-
-    stats = {}
-    report = {}
-
-    test_dataloaders = datamodule.test_dataloader()
-
-    for dts_name, dataloader in test_dataloaders.items():
-        # iterate all videos
-        auc_calc = AUROC(task="BINARY", num_classes=2)
-        acc_calc = Accuracy(task="BINARY", num_classes=2)
-        dataset = dataloader.dataset
-        dts_stats = {}
-
-        # perform ddp prediction
-        batch_results = trainer.predict(
-            model=model,
-            dataloaders=[dataloader]
-        )
-
-        gathered_results = [None] * torch.distributed.get_world_size()
-        torch.distributed.all_gather_object(gathered_results, batch_results)
-        torch.distributed.barrier()
-
-        if (trainer.is_global_zero):
-            # fetch predict results and aggregate.
-            for batch_results in gathered_results:
-                for batch_result in batch_results:
-                    probs = batch_result["probs"]
-                    names = batch_result["names"]
-                    y = batch_result["y"]
-                    for prob, label, name in zip(probs, y, names):
-                        if (not name in dts_stats):
-                            dts_stats[name] = StatsRecorder(label)
-                        dts_stats[name].update(prob, label)
-
-            # compute the average probability.
-            for k in dts_stats:
-                dts_stats[k] = dts_stats[k].compute()
-
-            # add straying videos into metric calculation
-            for k, v in dataset.stray_videos.items():
-                dts_stats[k] = dict(
-                    label=v,
-                    prob=0.5,
-                    stray=1
-                )
-
-            # compute the metric scores
-            dataset_labels = []
-            dataset_probs = []
-            for v in dts_stats.values():
-                dataset_labels.append(v["label"])
-                dataset_probs.append(v["prob"])
-            dataset_labels = torch.tensor(dataset_labels)
-            dataset_probs = torch.tensor(dataset_probs)
-            accuracy = acc_calc(dataset_probs, dataset_labels).item()
-            roc_auc = auc_calc(dataset_probs, dataset_labels).item()
-            accuracy = round(accuracy, 3)
-            roc_auc = round(roc_auc, 3)
-            logging.info(f'[{dts_name}] accuracy: {accuracy}, roc_auc: {roc_auc}')
-            stats[dts_name] = dts_stats
-            report[dts_name] = {
-                "accuracy": accuracy,
-                "roc_auc": roc_auc
-            }
-
-    if (trainer.is_global_zero):
-        # save report and stats.
-        with open(path.join(cfg_dir, f'report_{timestamp}.json'), "w") as f:
-            json.dump(report, f, sort_keys=True, indent=4, separators=(',', ': '))
-
-        with open(path.join(cfg_dir, f'stats_{timestamp}.pickle'), "wb") as f:
-            pickle.dump(stats, f)
-
-    return report
-
-
-if __name__ == "__main__":
-    configure_logging()
-
-    params = parse_args()
-
-    cli = ODLightningCLI(
-        run=False,
-        trainer_class=ODTrainer,
-        save_config_callback=None,
-        parser_kwargs={
-            "parser_mode": "omegaconf"
-        },
-        auto_configure_optimizers=False,
-        seed_everything_default=1019,
-        args=[
-            '-c', params.model_cfg_path,
-            '-c', params.data_cfg_path,
-            '--trainer.logger=null',
-            f'--trainer.devices={params.devices}',
-            f'--trainer.precision={params.precision}',
-        ],
-    )
-
-    cfg_dir = os.path.split(params.model_cfg_path)[0]
-    ckpt_path = params.model_ckpt_path
-    notes = params.notes
-
-    inference_driver(
-        cli=cli,
-        cfg_dir=cfg_dir,
-        ckpt_path=ckpt_path,
-        notes=notes
-    )

video/dfd-fcg/model_code/main.py

@@ -1,123 +0,0 @@
-import os
-import gc
-import json
-import torch
-import logging
-import warnings
-import lightning.pytorch as pl
-
-from lightning.pytorch.utilities import rank_zero_only
-from src.utility.builtin import ODTrainer, ODLightningCLI
-from inference import inference_driver
-torch.set_float32_matmul_precision('high')
-
-
-def configure_logging():
-    logging_fmt = "[%(levelname)s][%(filename)s:%(lineno)d]: %(message)s"
-    logging.basicConfig(level="INFO", format=logging_fmt)
-    warnings.filterwarnings(action="ignore")
-
-    # disable warnings from the xformers efficient attention module due to torch.user_deterministic_algorithms(True,warn_only=True)
-    warnings.filterwarnings(
-        action="ignore",
-        message=".*efficient_attention_forward_cutlass.*",
-        category=UserWarning
-    )
-
-    # logging.basicConfig(level="DEBUG", format=logging_fmt)
-
-
-def configure_cli():
-    return ODLightningCLI(
-        run=False,
-        trainer_class=ODTrainer,
-        save_config_kwargs={
-            'config_filename': 'setting.yaml'
-        },
-        auto_configure_optimizers=True,
-        seed_everything_default=1019
-    )
-
-
-def inference(cli):
-    # inference the best model
-    cfg_dir = cli.trainer.log_dir
-    ckpt_path = cli.trainer.checkpoint_callback.best_model_path
-
-    results = inference_driver(
-        cli=cli,
-        cfg_dir=cfg_dir,
-        ckpt_path=ckpt_path,
-    )
-
-    # log inference results
-    cli.trainer.logger.experiment.log(
-        {
-            "/".join(["infer", dts_name, metric]): value
-            for dts_name, metrics in results.items()
-            for metric, value in metrics.items()
-        },
-        commit=True
-    )
-
-    return results
-
-
-def cli_main():
-    # logging configuration
-    configure_logging()
-
-    # initialize cli
-    cli = configure_cli()
-
-    # update experiment notes
-    cli.trainer.logger.experiment.notes = cli.config.notes
-    cli.trainer.logger.experiment.save()
-
-    # monitor model gradient and parameter histograms
-    # (this severely slow down the training speed)
-    # cli.trainer.logger.experiment.watch(cli.model, log='all', log_graph=False)
-
-    # load & configure datasets
-    cli.datamodule.affine_model(cli.model)
-    cli.datamodule.affine_trainer(cli.trainer)
-
-    # determine the purpose of the given checkpoint
-    cont_ckpt_path = None
-    if not cli.config.ckpt_path is None:
-        if cli.config.ckpt_mode == "cont":
-            cont_ckpt_path = cli.config.ckpt_path
-        elif cli.config.ckpt_mode == "tune":
-            cli.model.load_state_dict(torch.load(cli.config.ckpt_path)["state_dict"])
-        else:
-            raise NotImplementedError()
-
-    # run
-    cli.trainer.fit(
-        cli.model,
-        datamodule=cli.datamodule,
-        ckpt_path=cont_ckpt_path
-    )
-
-    # after training:
-    # 1. unwatch model
-    # cli.trainer.logger.experiment.unwatch(cli.model)
-    # 2. save the config
-    cli.trainer.logger.experiment.save(
-        glob_str=os.path.join(cli.trainer.log_dir, 'setting.yaml'),
-        base_path=cli.trainer.log_dir,
-        policy="now"
-    )
-
-    gc.collect()
-    torch.cuda.empty_cache()
-
-    # inference the best model.
-    scores = inference(cli=cli)
-
-    # finally
-    cli.trainer.logger.experiment.finish()
-
-
-if __name__ == "__main__":
-    cli_main()

video/dfd-fcg/model_code/misc/20words_mean_face.npy

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:dbf68b2044171e1160716df7c53e8bbfaa0ee8c61fb41171d04cb6092bb81422
-size 1168

video/dfd-fcg/model_code/misc/L14_real_semantic_patches_v4_2000.pickle

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:438555d515700f0fdc4a1f2af2ca620d879fa7c91ca7018579e23aae94f0f72a
-size 2100301

video/dfd-fcg/model_code/readme.md

@@ -1,275 +0,0 @@
-<p align="center">
-  <h1 align="center">[CVPR'25] Towards More General Video-based Deepfake Detection through Facial Component Guided Adaptation for Foundation Model (DFD-FCG)</h1>
-
-  <p align="center">
-    <a href="https://github.com/ODD2"><strong>Yue-Hua Han</strong></a>
-    <sup>1,3,4</sup>
-    &nbsp;&nbsp;
-    <a href="https://github.com/Teddy12155555"><strong>Tai-Ming Huang</strong></a>
-    <sup>1,3,4</sup>
-    &nbsp;&nbsp;
-    <a href="https://scholar.google.com/citations?user=nnzQtDAAAAAJ&hl=zh-TW"><strong>Kai-Lung Hua</strong></a>
-    <sup>2,4</sup>
-    &nbsp;&nbsp;
-    <a href="https://scholar.google.com.au/citations?user=3x9KITUAAAAJ&hl=en"><strong> Jun-Cheng Chen</strong></a>
-    <sup>1</sup>
-    <br>
-    <!-- <sup>1</sup>Academia Sinica&nbsp;
-    <sup>2</sup>Microsoft&nbsp;
-    <sup>3</sup>National Taiwan University&nbsp;
-    <br>
-    <sup>4</sup>National Taiwan University of Science and Technology&nbsp; -->
-    <sup>1</sup>Academia Sinica,</span>&nbsp;
-    <sup>2</sup>Microsoft,</span>&nbsp;
-    <sup>3</sup>National Taiwan University,</span>&nbsp;
-    <br>
-    <sup>4</sup>National Taiwan University of Science and Technology</span>&nbsp;
-    <br>
-    <a href='https://arxiv.org/abs/2404.05583'><img src='https://img.shields.io/badge/ArXiv-2404.05583-red'></a>&nbsp;
-    <img src="assets/teaser.png">
-  </p>
-</p>
-
-## 🥇Abstract
-<div style="text-align: justify">  
-Generative models have enabled the creation of highly realistic facial-synthetic images, raising significant concerns due to their potential for misuse. Despite rapid advancements in the field of deepfake detection, developing efficient approaches to leverage foundation models for improved generalizability to unseen forgery samples remains challenging. To address this challenge, we propose a novel side-network-based decoder that extracts spatial and temporal cues using the CLIP image encoder for generalized video-based Deepfake detection. Additionally, we introduce Facial Component Guidance (FCG) to enhance spatial learning generalizability by encouraging the model to focus on key facial regions. By leveraging the generic features of a vision-language foundation model, our approach demonstrates promising generalizability on challenging Deepfake datasets while also exhibiting superiority in training data efficiency, parameter efficiency, and model robustness.
-</div>
-
-## 📝TODOs
-  - [x] Training + Evaluation Code
-  - [x] Model Weights
-  - [x] Inference Code
-  - [ ] HeyGen Evaluation Dataset
-
-
-## 🙌News
-  - June 08: We have released the [model checkpoint](https://drive.google.com/file/d/1ydD5rnaaF0i2zLE7NidLtAhjonHoVQOk/view?usp=sharing) and provided inference code for single videos! Checkout [this section](#inference---demo-video) for further details!
-
-## 🚀Installation
-```shell
-# conda environment
-conda env create -f environment.yml
-```
-
-## 📂Dataset Structure
-The structure of the **pre-processed datasets** for our project, the video files (*.avi) have been processed to only retain the aligned face. We use soft-links **(ln -s)** to manage and link the folders containing pre-processed videos on different drives.
-```shell
-datasets
-├── cdf
-│   ├── FAKE
-│   │   └── videos
-│   │         └── *.avi
-│   ├── REAL
-│   │   └── videos
-│   │         └── *.avi
-│   └── csv_files
-│       ├── test_fake.csv
-│       └── test_real.csv
-├── dfdc
-│   ├── csv_files
-│   │   └── test.csv
-│   └── videos
-├── dfo
-│   ├── FAKE
-│   │   └── videos
-│   │         └── *.avi
-│   ├── REAL
-│   │   └── videos
-│   │         └── *.avi
-│   └── csv_files
-│       ├── test_fake.csv
-│       └── test_real.csv
-├── ffpp
-│   ├── DF
-│   │   ├── c23
-│   │   │   └── videos
-│   │   │         └── *.avi
-│   │   ├── c40
-│   │   │   └── videos
-│   │   │         └── *.avi
-│   │   └── raw
-│   │       └── videos
-│   │   │         └── *.avi
-│   ├── F2F ...
-│   ├── FS ...
-│   ├── FSh ...
-│   ├── NT ...
-│   ├── real ...
-│   └── csv_files
-│       ├── test.json
-│       ├── train.json
-│       └── val.json
-|   
-└── robustness
-    ├── BW
-    │   ├── 1
-    │   │   ├── DF
-    │   │   │   └── c23
-    │   │   │       └── videos
-    │   │   │             └── *.avi
-    │   │   ├── F2F ...
-    │   │   ├── FS ...
-    │   │   ├── FSh ...
-    │   │   ├── NT ...
-    │   │   ├── real ...
-    │   │   │
-    │   │   └── csv_files
-    │   │       ├── test.json
-    │   │       ├── train.json
-    │   │       └── val.json
-    │   │   
-    │   │   
-    │   │   
-    .   .
-    .   .
-    .   .
-```
-
-
-## 🔧Dataset Pre-processing
-### Generic Pre-processing
-This phase performs the required pre-processing for our method, this includes *facial alignment (using the mean face from LRW)* and *facial cropping*.
-```bash
-# First, fetch all the landmarks & bboxes of the video frames.
-python -m src.preprocess.fetch_landmark_bbox \ 
---root-dir="/storage/FaceForensicC23" \ # The root folder of the dataset
---video-dir="videos" \  # The root folder of the videos
---fdata-dir="frame_data" \ # The folder to save the extracted frame data
---glob-exp="*/*" \  # The glob expression to search through the root video folder
---split-num=1 \ # Split the dataset into several parts for parallel process.
---part-num=1 \ # The part of dataset to process  for parallel process.
---batch=1 \ # The batch size for the 2D-FAN face data extraction. (suggestion: 1)
---max-res=800  # The maximum resolution for either side of the image
-
-# Then, crop all the faces from the original videos.
-python -m src.preprocess.crop_main_face \ 
---root-dir="/storage/FaceForensicC23/" \ # The root folder of the dataset
---video-dir="videos" \  # The root folder of the videos
---fdata-dir="frame_data" \ # The folder to fetch the frame data for landmarks and bboxes
---glob-exp="*/*" \  # The glob expression to search through the root video folder
---crop-dir="cropped" \ # The folder to save the cropped videos
---crop-width=150 \ # The width for the cropped videos
---crop-height=150 \ # The height for the cropped videos
---mean-face="./misc/20words_mean_face.npy" # The mean face for face aligned cropping. 
---replace \ # Control whether to replace existing cropped videos
---workers=1 # Number of works to perform parallel process (default: cpu / 2 )
-```
-
-
-### Robustness Pre-processing
-This phase requires the pre-processed facial landmarks to perform facial cropping, please refer to the **Generic Pre-processing** for further detail.
-```bash
-# First, we add perturbation to all the videos.
-python -m src.preprocess.phase1_apply_all_to_videos \ 
---dts-root="/storage/FaceForensicC23" \ # The root folder of the dataset
---vid-dir="videos" \  # The root folder of the videos
---rob-dir="robustness" \ # The folder to save the perturbed videos
---glob-exp="*/*.mp4" \  # The glob expression to search through the root video folder
---split=1 \ # Split the dataset into several parts for parallel process.
---part=1 \ # The part of dataset to process  for parallel process.
---workers=1  # Number of works to perform parallel process (default: cpu / 2 )
-
-# Then, crop all the faces from the perturbed videos.
-python -m src.preprocess.phase2_face_crop_all_videos \ 
-(setup/run/clean) # the three phase operations
---root-dir="/storage/FaceForensicC23/" \ # The root folder of the dataset
---rob-dir="videos" \  # The root folder of the robustness videos
---fd-dir="frame_data" \ # The folder to fetch the frame data for landmarks and bboxes
---glob-exp="*/*/*/*.mp4" \  # The glob expression to search through the root video folder
---crop-dir="cropped_robust" \ # The folder to save the cropped videos
---mean-face="./misc/20words_mean_face.npy" \ # The mean face for face aligned cropping. 
---workers=1  # Number of works to perform parallel process (default: cpu / 2 )
-```
-
-## 🤖Training & Evaluation
-### Training - Preset Settings 
-In `./scripts`,  scripts are provided to start the training process for the settings mentioned in our paper. 
-These settings are configured to run on a cluster with `V100*4`.
-```bash
-bash ./scripts/model/ffg_l14.sh # begin training process
-```
-### Training - Custom Settings
-Our project is built on `pytorch-lightning (2.2.0)`, please refer the  [official manual](https://lightning.ai/docs/pytorch/2.2.0/common/trainer.html#trainer-class-api) and adjust the following files for advance configurations:
-```bash
-./configs/base.yaml # major training settings (e.g. epochs, optimizer, batch size, mixed-precision ...)
-./configs/data.yaml # settings for the training & validation dataset
-./configs/inference.yaml # settings for the evaluation dataset (extension of data.yaml)
-./configs/logger.yaml # settings for the WandB logger
-./configs/clip/L14/ffg.yaml # settings for the main model 
-./configs/test.yaml # settings for debugging (offline logging, small batch size, short epochs ...)
-```
-The following command starts the training process with the provided settings:
-```bash
-# For debugging, add '--config configs/test.yaml' after the '--config configs/clip/L14/ffg.yaml'.
-python main.py \
---config configs/base.yaml \
---config configs/clip/L14/ffg.yaml
-
-# Fine-grained control is supported with the pytorch-lightning-cli.
-python main.py \
---config configs/base.yaml \
---config configs/clip/L14/ffg.yaml \
---optimizer.lr=1e-5 \
---trainer.max_epochs=10 \
---data.init_args.train_datamodules.init_args.batch_size=5
-```
-### Evaluation - Standard
-To perform evaluation on datasets, run the following command:
-```bash
-python inference.py \
-"logs/fcg_l14/setting.yaml" \ # model settings
-"./configs/inference.yaml" \ # evaluation dataset settings
-"logs/fcg_l14/checkpoint.ckpt" \ # model checkpoint
-"--devices=4" # number of devices to compute in parallel
-```
-### Evaluation - Robustness
-We provide tools in `./scripts/tools/` to simplify the robustness evaluation task: `create-robust-config.sh` creates an evaluation config for each perturbation types and `inference-robust.sh` runs through all the datasets with the specified model.
-
-## 😎Inference - Demo Video
-To run inference on a single video with an indicator, please download our model checkpoint and execute the following commands:
-```bash
-# Pre-Processing: fetch facial landmark and bounding box
-python -m src.preprocess.fetch_landmark_bbox \
---root-dir="./resources" \
---video-dir="videos" \
---fdata-dir="frame_data" \
---glob-exp="*"
-# Pre-Processing: crop out the facial regions
-python -m src.preprocess.crop_main_face \
---root-dir="./resources" \
---video-dir="videos" \
---fdata-dir="frame_data" \
---crop-dir="cropped" \
---glob-exp="*"
-# Main Process
-python -m demo \
-"checkpoint/setting.yaml" \ # the model setting of the checkpoint
-"checkpoint/weights.ckpt" \ # the model weights of the checkpoint
-"resources/videos/000_003.mp4" \ # the video to process
---out_path="test.avi" \ # the output path of the processed video
---threshold=0.5 \ # the threshold for the real/fake indicator
---batch_size=30 # the input batch size of the model (~10G VRAM when batch_size=30 )
-```
-The following is a sample frame from the processed video:
-<p align="center">
-    <img src="assets/demo.png">
-</p>
-
-
-<!-- ## 🔥 Inference -->
-## 🔗 BibTeX
-If you find our efforts helpful, please cite our paper and leave a star for further updates!
-```bibtex
-
-@inproceedings{cvpr25_dfd_fcg,
-      title={Towards More General Video-based Deepfake Detection through Facial Component Guided Adaptation for Foundation Model},
-      author={Yue-Hua Han, Tai-Ming Huang, Kai-Lung Hua, Jun-Cheng Chen},
-      booktitle={Proceedings of the Conference on Computer Vision and Pattern Recognition (CVPR)},
-      year={2025}
-}
-```
-
-
-## 📭 Contact
-The provided code and weights are only available for research purpose only. 
-If you have further questions (including commercial use), please contact [Dr. Jun-Cheng Chen](pullpull@citi.sinica.edu.tw).

video/dfd-fcg/model_code/resources/videos/000.mp4

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:0e51b9b46acc7eaeca62022e1b869541406d2ca887cd401dcc03cc5db28bc8ed
-size 894312

video/dfd-fcg/model_code/resources/videos/000_003.mp4

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:6e83d0e0c51a0f7cdf2db48f796617db43148b3390011555dfa758f2b7ab811b
-size 891837

video/dfd-fcg/model_code/scripts/ablation/ffg.sh

@@ -1,6 +0,0 @@
-python -m main \
---config configs/base.yaml \
---config configs/models/svl.yaml \
---model.init_args.num_synos=4 \
---config configs/generic/inference.yaml \
---notes="no ffg guidance"