kitti-flow2012.py

import datasets
import os
from PIL import Image

_KITTI2012_URL = "https://s3.eu-central-1.amazonaws.com/avg-kitti/data_stereo_flow.zip"

class KITTIFLow2012(datasets.GeneratorBasedBuilder):
    """KITTI Flow 2012 dataset with grayscale/color image sequences, optical flow ground truth, and calibration."""

    VERSION = datasets.Version("1.0.0")

    def _info(self):
        return datasets.DatasetInfo(
            description=(
                "KITTI Flow 2012 dataset: contains grayscale and color stereo image sequences "
                "captured at two consecutive timepoints, along with sparse optical flow ground truth and calibration files. "
                "This dataset is widely used for benchmarking optical flow estimation methods in realistic driving scenarios."
            ),
            features=datasets.Features(
                {
                    "ImageGray_left": datasets.Sequence(datasets.Image()),
                    "ImageGray_right": datasets.Sequence(datasets.Image()),
                    "ImageColor_left": datasets.Sequence(datasets.Image()),
                    "ImageColor_right": datasets.Sequence(datasets.Image()),
                    "calib": {
                        "P0": datasets.Sequence(datasets.Value("float32")),
                        "P1": datasets.Sequence(datasets.Value("float32")),
                        "P2": datasets.Sequence(datasets.Value("float32")),
                        "P3": datasets.Sequence(datasets.Value("float32")),
                    },
                    "disp_noc": datasets.Image(),
                    "disp_occ": datasets.Image(),
                    "disp_refl_noc": datasets.Image(),
                    "disp_refl_occ": datasets.Image(),
                    "flow_noc": datasets.Image(),
                    "flow_occ": datasets.Image(),
                }
            ),
            homepage="http://www.cvlibs.net/datasets/kitti/eval_stereo_flow.php?benchmark=flow",
            license="CC BY-NC-SA 3.0",
            citation="""@inproceedings{Geiger2012CVPR,
  author = {Andreas Geiger and Philip Lenz and Raquel Urtasun},
  title = {Are we ready for Autonomous Driving? The KITTI Vision Benchmark Suite},
  booktitle = {Conference on Computer Vision and Pattern Recognition (CVPR)},
  year = {2012}
}""",
        )

    def _split_generators(self, dl_manager):
        archive_path = dl_manager.download_and_extract(_KITTI2012_URL)
        train_path = os.path.join(archive_path, "training")
        test_path = os.path.join(archive_path, "testing")
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={"base_path": train_path, "split": "training"},
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={"base_path": test_path, "split": "testing"},
            ),
        ]

    def _generate_examples(self, base_path, split):
        image_0_path = os.path.join(base_path, "image_0")
        image_1_path = os.path.join(base_path, "image_1")
        color_0_path = os.path.join(base_path, "colored_0")
        color_1_path = os.path.join(base_path, "colored_1")
        calib_path = os.path.join(base_path, "calib")

        if split == "training":
            disp_path = os.path.join(base_path, "disp_noc")
            disp_path_occ = os.path.join(base_path, "disp_occ")
            disp_path_refl_noc = os.path.join(base_path, "disp_refl_noc")
            disp_path_refl_occ = os.path.join(base_path, "disp_refl_occ")
            flow_noc_path = os.path.join(base_path, "flow_noc")
            flow_occ_path = os.path.join(base_path, "flow_occ")

        files = sorted(os.listdir(image_0_path))
        ids = sorted(set(f.split("_")[0] for f in files))

        for id_ in ids:
            example = {
                "ImageGray_left": [
                    Image.open(os.path.join(image_0_path, f"{id_}_10.png")),
                    Image.open(os.path.join(image_0_path, f"{id_}_11.png")),
                ],
                "ImageGray_right": [
                    Image.open(os.path.join(image_1_path, f"{id_}_10.png")),
                    Image.open(os.path.join(image_1_path, f"{id_}_11.png")),
                ],
                "ImageColor_left": [
                    Image.open(os.path.join(color_0_path, f"{id_}_10.png")),
                    Image.open(os.path.join(color_0_path, f"{id_}_11.png")),
                ],
                "ImageColor_right": [
                    Image.open(os.path.join(color_1_path, f"{id_}_10.png")),
                    Image.open(os.path.join(color_1_path, f"{id_}_11.png")),
                ],
                "calib": {"P0": [], "P1": [], "P2": [], "P3": []},
                "disp_noc": Image.open(os.path.join(disp_path, f"{id_}_10.png")) if split == "training" else None,
                "disp_occ": Image.open(os.path.join(disp_path_occ, f"{id_}_10.png")) if split == "training" else None,
                "disp_refl_noc": Image.open(os.path.join(disp_path_refl_noc, f"{id_}_10.png")) if split == "training" else None,
                "disp_refl_occ": Image.open(os.path.join(disp_path_refl_occ, f"{id_}_10.png")) if split == "training" else None,
                "flow_noc": Image.open(os.path.join(flow_noc_path, f"{id_}_10.png")) if split == "training" else None,
                "flow_occ": Image.open(os.path.join(flow_occ_path, f"{id_}_10.png")) if split == "training" else None,
            }

            calib_file = os.path.join(calib_path, f"{id_}.txt")
            with open(calib_file, "r") as f:
                lines = f.readlines()
                for line in lines:
                    key, value = line.strip().split(":", 1)
                    if key in ["P0", "P1", "P2", "P3"]:
                        example["calib"][key] = [float(x) for x in value.strip().split()]

            yield id_, example