put in final fix

video_std_manip.py

@@ -21,7 +21,7 @@ _CITATION = """\
 """
 
 _DESCRIPTION = """\
-This dataset is a collection of simple and traditional localized video manipulations, such as: splicing, color correction, contrast enhancement, bluring, and noise addition. The dataset is designed to be used for training and evaluating video manipulation detection models. We used this dataset to train the VideoFACT model, which is a deep learning model that uses attention, scene context, and forensic traces to detect a wide variety of video forgery types, i.e. splicing, editing, deepfake, inpainting. The dataset is divided into three parts: Video Camera Model Splicing (VCMS), Video Perceptually Visible Manipulation (VPVM), and Video Perceptually Invisible Manipulation (VPIM). Each part has a total of 2000 videos, each video is 1 second, or 30 frames, has a resolution of 1920 x 1080, and encoded using FFmpeg with the H.264 codec at CRF 23. Additionally, each part is splited into training, validation, and testing sets that consists of 1600, 100, 300 videos, respectively. More details about the dataset can be found in the paper. 
+This dataset is a collection of simple and traditional localized video manipulations, such as: splicing, color correction, contrast enhancement, bluring, and noise addition. The dataset is designed to be used for training and evaluating video manipulation detection models. We used this dataset to train the VideoFACT model, which is a deep learning model that uses attention, scene context, and forensic traces to detect a wide variety of video forgery types, i.e. splicing, editing, deepfake, inpainting. The dataset is divided into three parts: Video Camera Model Splicing (VCMS), Video Perceptually Visible Manipulation (VPVM), and Video Perceptually Invisible Manipulation (VPIM). Each part has a total of 4000 videos, each video is 1 second, or 30 frames, has a resolution of 1920 x 1080, and encoded using FFmpeg with the H.264 codec at CRF 23. Additionally, each part is splited into training, validation, and testing sets that consists of 3200, 200, 600 videos, respectively. More details about the dataset can be found in the paper. 
 """
 
 _HOMEPAGE = "https://github.com/ductai199x/videofact-wacv-2024"
@@ -41,7 +41,6 @@ class VideoStdManip(datasets.GeneratorBasedBuilder):
     """This dataset is a collection of simple and traditional localized video manipulations, such as: splicing, color correction, contrast enhancement, bluring, and noise addition. The dataset is divided into three parts: Video Camera Model Splicing (VCMS), Video Perceptually Visible Manipulation (VPVM), and Video Perceptually Invisible Manipulation (VPIM)."""
 
     VERSION = datasets.Version("1.0.0")
-    IN_MEMORY_MAX_SIZE = 1.0e10
 
     # This is an example of a dataset with multiple configurations.  
     # If you don't want/need to define several sub-sets in your dataset,
@@ -65,8 +64,8 @@ class VideoStdManip(datasets.GeneratorBasedBuilder):
     def _info(self):
         features = datasets.Features(
             {
-                "frames": datasets.Array4D(shape=(30, 1080, 1920, 3), dtype="uint8"),
-                "masks": datasets.Array3D(shape=(30, 1080, 1920), dtype="uint8"),
+                "vid_path": datasets.Value("string"),
+                "mask_path": datasets.Value("string"),
                 "label": datasets.ClassLabel(num_classes=2),
                 # These are the features of your dataset like images, labels ...
             }
@@ -137,16 +136,14 @@ class VideoStdManip(datasets.GeneratorBasedBuilder):
         for key, (label, vid_id) in enumerate(all_vid_ids):
             label = 0 if label == "real" else 1
             if label == 1:
-                frames = decord.VideoReader(open(os.path.join(part_dir, "manipulated", vid_id + ".mp4"), "rb"))[:]
-                masks = decord.VideoReader(open(os.path.join(part_dir, "mask", vid_id + ".mp4"), "rb"))[:]
+                vid_path = os.path.join(part_dir, "manipulated", vid_id + ".mp4")
+                mask_path = os.path.join(part_dir, "mask", vid_id + ".mp4")
             else:
-                frames = decord.VideoReader(open(os.path.join(part_dir, "original", vid_id + ".mp4"), "rb"))[:]
-                masks = torch.zeros_like(frames, dtype=torch.uint8)
-            
-            masks = ((masks.float().mean(3) / 255.0) > 0.5).to(torch.uint8)
+                vid_path = os.path.join(part_dir, "original", vid_id + ".mp4")
+                mask_path = ""
 
             yield key, {
-                "frames": frames,
-                "masks": masks,
+                "vid_path": vid_path,
+                "mask_path": mask_path,
                 "label": label,
             }