MPI-Sintel

http://sintel.is.tue.mpg.de/

root=data/sintel
mkdir -p ${root} && cd ${root}
f=MPI-Sintel-complete.zip
wget http://files.is.tue.mpg.de/sintel/${f}
unzip -q ${f}
cd ../..

__all__ = [
    'visualize',
]

from typing import Any, Iterable

T = dict[str, Any]


def visualize(dataset: Iterable[T]) -> None:
    from collections import defaultdict

    import cv2
    import numpy as np
    import tqdm

    import todd.tasks.optical_flow_estimation as ofe
    from todd.patches.cv2 import ColorMap, VideoWriter

    color_map = ColorMap(cv2.COLORMAP_JET)
    video_writers: defaultdict[str, VideoWriter] = \
        defaultdict(lambda: VideoWriter(12.))
    for data in tqdm.tqdm(dataset):
        id_: str = data['id_']
        of = ofe.OpticalFlow(data['of'])
        frame = np.concatenate([
            data['frame1'],
            of.to_color().numpy(),
            color_map(of.a),
            color_map(of.r),
            color_map(of.u),
            color_map(of.v),
        ])
        scene, _ = id_.split('/', 1)
        video_writers[scene].write(frame)
    for name, video_writer in video_writers.items():
        video_writer.dump(f'{name}.mp4')

import pathlib

import todd.tasks.optical_flow_estimation as ofe
from todd import Config
from todd.configs import PyConfig

dataset = ofe.datasets.SintelDataset(
    access_layer=Config(data_root='data/sintel'),
    pass_='final',  # nosec B106
)
PyConfig.load(
    pathlib.Path(__file__).parent / 'optical_flow.py',
).visualize(dataset)

data/sintel/
├── training
│   ├── albedo|clean|final|flow_viz|invalid|occlusions
│   │   ├── alley_1
│   │   │   ├── frame_0001.png
│   │   │   └── ...
│   │   └── ...
│   └── flow
│       ├── alley_1
│       │   ├── frame_0001.flo
│       │   └── ...
│       └── ...
└── test
    └── clean|final
        ├── ambush_1
        │   ├── frame_0001.png
        │   └── ...
        └── ...