137 lines
5.1 KiB
Python
137 lines
5.1 KiB
Python
# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
|
# SPDX-License-Identifier: Apache-2.0
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
import os
|
|
from functools import wraps
|
|
import math
|
|
import torch
|
|
from PIL import Image
|
|
import torchvision
|
|
|
|
from chronoedit._ext.imaginaire.utils import log
|
|
_IMAGE_EXTENSIONS = [".png", ".jpg", ".jpeg", "webp"]
|
|
|
|
def resize_input(video: torch.Tensor, resolution: list[int]):
|
|
r"""
|
|
Resizes and crops the input video tensor while preserving aspect ratio.
|
|
|
|
The video is first resized so that the smaller dimension matches the target resolution,
|
|
preserving the aspect ratio. Then, it's center-cropped to the target resolution.
|
|
|
|
Args:
|
|
video (torch.Tensor): Input video tensor of shape (T, C, H, W).
|
|
resolution (list[int]): Target resolution [H, W].
|
|
|
|
Returns:
|
|
torch.Tensor: Resized and cropped video tensor of shape (T, C, target_H, target_W).
|
|
"""
|
|
|
|
orig_h, orig_w = video.shape[2], video.shape[3]
|
|
target_h, target_w = resolution
|
|
|
|
scaling_ratio = max((target_w / orig_w), (target_h / orig_h))
|
|
resizing_shape = (int(math.ceil(scaling_ratio * orig_h)), int(math.ceil(scaling_ratio * orig_w)))
|
|
video_resized = torchvision.transforms.functional.resize(video, resizing_shape)
|
|
video_cropped = torchvision.transforms.functional.center_crop(video_resized, resolution)
|
|
return video_cropped
|
|
def read_and_process_image(img_path: str, resolution: list[int], num_video_frames: int, resize: bool = True):
|
|
"""
|
|
Reads an image, converts it to a video tensor, and processes it for model input.
|
|
|
|
The image is loaded, converted to a tensor, and replicated to match the
|
|
`num_video_frames`. It's then optionally resized and permuted to the
|
|
standard video format (B, C, T, H, W).
|
|
|
|
Args:
|
|
img_path (str): Path to the input image file.
|
|
resolution (list[int]): Target resolution [H, W] for resizing.
|
|
num_video_frames (int): The number of frames the output video tensor should have.
|
|
resize (bool, optional): Whether to resize the image to the target resolution. Defaults to True.
|
|
|
|
Returns:
|
|
torch.Tensor: Processed video tensor of shape (1, C, T, H, W).
|
|
|
|
Raises:
|
|
ValueError: If the image extension is not one of the supported types.
|
|
"""
|
|
ext = os.path.splitext(img_path)[1]
|
|
if ext.lower() not in _IMAGE_EXTENSIONS:
|
|
raise ValueError(f"Invalid image extension: {ext}")
|
|
|
|
# Read the image
|
|
img = Image.open(img_path)
|
|
|
|
# Convert to tensor
|
|
img = torchvision.transforms.functional.to_tensor(img)
|
|
# Create a video tensor by repeating the first frame
|
|
vid_input = img.unsqueeze(0) # Add temporal dimension T=1
|
|
|
|
# Repeat the first frame to match the desired number of video frames
|
|
# Note: The actual content for frames > 0 will be generated by the model.
|
|
vid_input = torch.cat([vid_input, torch.zeros_like(vid_input).repeat(num_video_frames - 1, 1, 1, 1)], dim=0)
|
|
vid_input = (vid_input * 255.0).to(torch.uint8) # Convert to uint8 range if needed (might depend on model)
|
|
if resize:
|
|
# Resize and crop to the target resolution
|
|
vid_input = resize_input(vid_input, resolution)
|
|
|
|
# Convert to {B, C, T, H, W} format expected by the model
|
|
vid_input = vid_input.unsqueeze(0).permute(0, 2, 1, 3, 4) # Add batch dim B=1 and permute
|
|
return vid_input
|
|
|
|
class sync_timer:
|
|
"""
|
|
Synchronized timer to count the inference time of `nn.Module.forward` or else.
|
|
set env var SYNC_TIMER=1 to enable logging!
|
|
|
|
Example as context manager:
|
|
```python
|
|
with timer('name'):
|
|
run()
|
|
```
|
|
|
|
Example as decorator:
|
|
```python
|
|
@timer('name')
|
|
def run():
|
|
pass
|
|
```
|
|
"""
|
|
|
|
def __init__(self, name=None, flag_env="SYNC_TIMER"):
|
|
self.name = name
|
|
self.flag_env = flag_env
|
|
|
|
def __enter__(self):
|
|
if os.environ.get(self.flag_env, "0") == "1":
|
|
self.start = torch.cuda.Event(enable_timing=True)
|
|
self.end = torch.cuda.Event(enable_timing=True)
|
|
self.start.record()
|
|
return lambda: self.time
|
|
|
|
def __exit__(self, exc_type, exc_value, exc_tb):
|
|
if os.environ.get(self.flag_env, "0") == "1":
|
|
self.end.record()
|
|
torch.cuda.synchronize()
|
|
self.time = self.start.elapsed_time(self.end)
|
|
if self.name is not None:
|
|
log.info(f"{self.name} takes {self.time/1000:.4f}s", rank0_only=False)
|
|
|
|
def __call__(self, func):
|
|
@wraps(func)
|
|
def wrapper(*args, **kwargs):
|
|
with self:
|
|
result = func(*args, **kwargs)
|
|
return result
|
|
return wrapper
|