add dataset

.gitattributes

@@ -59,3 +59,15 @@ mixkit.00.tar filter=lfs diff=lfs merge=lfs -text
 pexels.00.tar filter=lfs diff=lfs merge=lfs -text
 pexels.01.tar filter=lfs diff=lfs merge=lfs -text
 pexels.02.tar filter=lfs diff=lfs merge=lfs -text
+pixabay.01.tar filter=lfs diff=lfs merge=lfs -text
+pixabay.03.tar filter=lfs diff=lfs merge=lfs -text
+pixabay.07.tar filter=lfs diff=lfs merge=lfs -text
+pixabay.08.tar filter=lfs diff=lfs merge=lfs -text
+pixabay.11.tar filter=lfs diff=lfs merge=lfs -text
+pixabay.09.tar filter=lfs diff=lfs merge=lfs -text
+pixabay.10.tar filter=lfs diff=lfs merge=lfs -text
+pixabay.00.tar filter=lfs diff=lfs merge=lfs -text
+pixabay.02.tar filter=lfs diff=lfs merge=lfs -text
+pixabay.04.tar filter=lfs diff=lfs merge=lfs -text
+pixabay.05.tar filter=lfs diff=lfs merge=lfs -text
+pixabay.06.tar filter=lfs diff=lfs merge=lfs -text

mixkit.00.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43273d48d9a2c7d47726318e32a61acd8fbbbc75855ca8080e3fad6f1b1b6e05
+size 803870720

pexels.00.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:91028024dd214584f182ef0ef2164e4670cd593f8434f3c9a281ee1cb77ccfec
+size 3054643200

pexels.01.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c1d96cfa7c85a67d07a3ab50bbf4a8a3afa9242ef0df48a97421b360054a367d
+size 3054684160

pexels.02.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d9f9d164bccaffcf1417fbfae665c7bc6294a59a9a5a532913ce4666a12a6851
+size 1350881280

pixabay.00.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0649c4ec78c3fc1c05d027c5c26bcab56089f25c471c809f767fad52ae31d5a4
+size 3057971200

pixabay.01.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c6da266c18ae8cc2f79561dda254ff55ea89bafffb03e9d18e434d41b431abe
+size 3053998080

pixabay.02.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3dbe15983a893a69b0d672270da0623ee31d294af8b1d0c4b46f42db89d829a
+size 3055452160

pixabay.03.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd9a26844bc0c1a563a24392b26524de5bae7f44adc104ceaebb995cbb8b2f63
+size 3055431680

pixabay.04.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5493e87d9badab0f26758a392a4ff8502f332acd71f8b378a22390c4223baf13
+size 3054643200

pixabay.05.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca8a07b3a59ec0297441f8bbc33d6656e2ca20107f927bdeccdd146c462cb771
+size 3052533760

pixabay.06.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09564c3253829a287fbd4425b190bf854738c13b907ae95ddb8b7b0754f424ba
+size 3052984320

pixabay.07.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:adfa63148ffbed018e8efec0e122a8411a5d59d09bf0926148a20e3a4a9eabc6
+size 3052769280

pixabay.08.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d13c9ea32c82e904bcf74726ebb154d49ae0bf604f02119b1eddf9859bcbd26a
+size 3053219840

pixabay.09.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ddad46f28e19cb9e33ec8e17a082958950f57fe8aa0e3dbdbc7af854a7d00ae1
+size 3056230400

pixabay.10.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43c37d37b457820d76e8781ec8f336ae6864c6ab59c61b61d1e80c3b58bb76c2
+size 3054008320

pixabay.11.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e49d160a33355ac058ec3bba753c0802fae3eeb76170b9cba599e61c8e3f2876
+size 1696389120

run.sh

@@ -0,0 +1,34 @@
+set -ex
+
+MODE=$1
+DATASET=${2:-mixkit}
+echo "Running in mode: $MODE"
+
+if [ "$MODE" == "sample_all" ]; then
+    conda run --no-capture-output -n torch python sample.py --root_dir mixkit --size 512
+    rm -rf mixkit
+    conda run --no-capture-output -n torch python sample.py --root_dir pexels --size 512
+    rm -rf pexels
+    conda run --no-capture-output -n torch python sample.py --root_dir pixabay --size 512 
+    rm -rf pixabay
+elif [ "$MODE" == "sample" ]; then
+    if [ "$DATASET" == "mixkit" ]; then
+	unpigz -c mixkit.tar.gz | tar -xvf -
+    elif [ "$DATASET" == "pexels" ]; then
+    	cat pexels.tar.gz.0* | unpigz -c - | tar -xvf - 
+    elif [ "$DATASET" == "pixabay" ]; then
+	cat pixabay.tar.gz.{00..24} | unpigz -c - | tar -xvf -
+    fi
+    rm -rf $DATASET.tar.gz*
+    conda run --no-capture-output -n torch python sample.py --root_dir $DATASET --size 512
+    rm -rf $DATASET
+elif [ "$MODE" == "sample_files"]; then
+    conda run --no-capture-output -n torch python sample.py --root_dir mixkit --size 512 --output_format files
+    rm -rf mixkit
+    conda run --no-capture-output -n torch python sample.py --root_dir pexels --size 512 --output_format files
+    rm -rf pexels
+    conda run --no-capture-output -n torch python sample.py --root_dir pixabay --size 512 --output_format files
+    rm -rf pixabay
+else
+    echo "Invalid mode"
+fi

sample.py

@@ -0,0 +1,70 @@
+import argparse
+from torch.utils.data import Dataset, DataLoader
+from glob import glob
+import os, os.path as osp
+from tqdm import tqdm
+import webdataset as wds
+
+from kn_util.data.video import read_frames_decord, get_frame_indices, fill_temporal_param, save_video_imageio, array_to_video_bytes
+from kn_util.data.transforms.video import Compose, Resize, ToStackedArray, CenterCrop
+from kn_util.data.transforms.video.functional import split_array
+
+
+class VideoDataset(Dataset):
+    def __init__(self, root_dir, fps=8, size=256):
+        self.transform = Compose(
+            [
+                Resize(size),
+                ToStackedArray(),
+            ]
+        )
+        self.video_files = glob(osp.join(root_dir, "**/*.mp4"))
+        self.fps = fps
+
+    def __getitem__(self, idx):
+        video_file = self.video_files[idx]
+
+        frames = read_frames_decord(video_file, fps=self.fps, output_format="thwc")
+        frames = split_array(frames)
+        frames = self.transform(frames)
+        return video_file, frames
+
+    def __len__(self):
+        return len(self.video_files)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--root_dir", type=str)
+    parser.add_argument("--output_dir", type=str, default="outputs")
+    parser.add_argument("--fps", type=int, default=8)
+    parser.add_argument("--size", type=int, default=256)
+    parser.add_argument("--output_format", type=str, default="tar")
+    args = parser.parse_args()
+
+    dataset = VideoDataset(root_dir=args.root_dir, fps=args.fps, size=args.size)
+    dataloader = DataLoader(
+        dataset,
+        batch_size=1,
+        shuffle=False,
+        collate_fn=lambda x: x[0],
+        num_workers=32,
+        pin_memory=True,
+    )
+
+    dataset_name = args.root_dir.strip("/")
+    tar_path = osp.join(args.output_dir, f"{dataset_name}.%02d.tar")
+    os.makedirs(args.output_dir, exist_ok=True)
+
+    if args.output_format == "files":
+        for video_path, frames in tqdm(dataloader):
+            output_path = osp.join(args.output_dir, video_path)
+            os.makedirs(osp.dirname(output_path), exist_ok=True)
+            save_video_imageio(frames, output_path, fps=args.fps, input_format="thwc")
+
+    if args.output_format == "tar":
+        writer = wds.ShardWriter(tar_path, maxcount=4000000000)
+        for video_path, frames in tqdm(dataloader):
+            video_filename = osp.basename(video_path).split(".")[0]
+            writer.write({"mp4": array_to_video_bytes(frames), "__key__": video_filename})
+        writer.close()