Upload 4 files

hyvideo/utils/data_utils.py

@@ -0,0 +1,15 @@
+import numpy as np
+import math
+
+
+def align_to(value, alignment):
+    """align hight, width according to alignment
+
+    Args:
+        value (int): height or width
+        alignment (int): target alignment factor
+
+    Returns:
+        int: the aligned value
+    """
+    return int(math.ceil(value / alignment) * alignment)

hyvideo/utils/file_utils.py

@@ -0,0 +1,70 @@
+import os
+from pathlib import Path
+from einops import rearrange
+
+import torch
+import torchvision
+import numpy as np
+import imageio
+
+CODE_SUFFIXES = {
+    ".py",  # Python codes
+    ".sh",  # Shell scripts
+    ".yaml",
+    ".yml",  # Configuration files
+}
+
+
+def safe_dir(path):
+    """
+    Create a directory (or the parent directory of a file) if it does not exist.
+
+    Args:
+        path (str or Path): Path to the directory.
+
+    Returns:
+        path (Path): Path object of the directory.
+    """
+    path = Path(path)
+    path.mkdir(exist_ok=True, parents=True)
+    return path
+
+
+def safe_file(path):
+    """
+    Create the parent directory of a file if it does not exist.
+
+    Args:
+        path (str or Path): Path to the file.
+
+    Returns:
+        path (Path): Path object of the file.
+    """
+    path = Path(path)
+    path.parent.mkdir(exist_ok=True, parents=True)
+    return path
+
+def save_videos_grid(videos: torch.Tensor, path: str, rescale=False, n_rows=1, fps=24):
+    """save videos by video tensor
+       copy from https://github.com/guoyww/AnimateDiff/blob/e92bd5671ba62c0d774a32951453e328018b7c5b/animatediff/utils/util.py#L61
+
+    Args:
+        videos (torch.Tensor): video tensor predicted by the model
+        path (str): path to save video
+        rescale (bool, optional): rescale the video tensor from [-1, 1] to  . Defaults to False.
+        n_rows (int, optional): Defaults to 1.
+        fps (int, optional): video save fps. Defaults to 8.
+    """
+    videos = rearrange(videos, "b c t h w -> t b c h w")
+    outputs = []
+    for x in videos:
+        x = torchvision.utils.make_grid(x, nrow=n_rows)
+        x = x.transpose(0, 1).transpose(1, 2).squeeze(-1)
+        if rescale:
+            x = (x + 1.0) / 2.0  # -1,1 -> 0,1
+        x = torch.clamp(x, 0, 1)
+        x = (x * 255).numpy().astype(np.uint8)
+        outputs.append(x)
+
+    os.makedirs(os.path.dirname(path), exist_ok=True)
+    imageio.mimsave(path, outputs, fps=fps)

hyvideo/utils/helpers.py

@@ -0,0 +1,40 @@
+import collections.abc
+
+from itertools import repeat
+
+
+def _ntuple(n):
+    def parse(x):
+        if isinstance(x, collections.abc.Iterable) and not isinstance(x, str):
+            x = tuple(x)
+            if len(x) == 1:
+                x = tuple(repeat(x[0], n))
+            return x
+        return tuple(repeat(x, n))
+    return parse
+
+
+to_1tuple = _ntuple(1)
+to_2tuple = _ntuple(2)
+to_3tuple = _ntuple(3)
+to_4tuple = _ntuple(4)
+
+
+def as_tuple(x):
+    if isinstance(x, collections.abc.Iterable) and not isinstance(x, str):
+        return tuple(x)
+    if x is None or isinstance(x, (int, float, str)):
+        return (x,)
+    else:
+        raise ValueError(f"Unknown type {type(x)}")
+
+
+def as_list_of_2tuple(x):
+    x = as_tuple(x)
+    if len(x) == 1:
+        x = (x[0], x[0])
+    assert len(x) % 2 == 0, f"Expect even length, got {len(x)}."
+    lst = []
+    for i in range(0, len(x), 2):
+        lst.append((x[i], x[i + 1]))
+    return lst

hyvideo/utils/preprocess_text_encoder_tokenizer_utils.py

@@ -0,0 +1,46 @@
+import argparse
+import torch
+from transformers import (
+    AutoProcessor,
+    LlavaForConditionalGeneration,
+)
+
+
+def preprocess_text_encoder_tokenizer(args):
+
+    processor = AutoProcessor.from_pretrained(args.input_dir)
+    model = LlavaForConditionalGeneration.from_pretrained(
+        args.input_dir,
+        torch_dtype=torch.float16,
+        low_cpu_mem_usage=True,
+    ).to(0)
+
+    model.language_model.save_pretrained(
+        f"{args.output_dir}"
+    )
+    processor.tokenizer.save_pretrained(
+        f"{args.output_dir}"
+    )
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--input_dir",
+        type=str,
+        required=True,
+        help="The path to the llava-llama-3-8b-v1_1-transformers.",
+    )
+    parser.add_argument(
+        "--output_dir",
+        type=str,
+        default="",
+        help="The output path of the llava-llama-3-8b-text-encoder-tokenizer."
+        "if '', the parent dir of output will be the same as input dir.",
+    )
+    args = parser.parse_args()
+
+    if len(args.output_dir) == 0:
+        args.output_dir = "/".join(args.input_dir.split("/")[:-1])
+
+    preprocess_text_encoder_tokenizer(args)