tomatotest.py

import io
from PIL import Image
from datasets import GeneratorBasedBuilder, DatasetInfo, Features, SplitGenerator, Value, Array2D, Split
import datasets
import numpy as np
import h5py
from huggingface_hub import HfFileSystem

class CustomConfig(datasets.BuilderConfig):
    def __init__(self, **kwargs):
        super(CustomConfig, self).__init__(**kwargs)
        self.dataset_type = kwargs.pop("name", "all")
_metadata_urls = {
    "train":"https://huggingface.co/datasets/XingjianLi/tomatotest/resolve/main/train.txt",
    "val":"https://huggingface.co/datasets/XingjianLi/tomatotest/resolve/main/val.txt"

}

class RGBSemanticDepthDataset(GeneratorBasedBuilder):
    BUILDER_CONFIGS = [
        CustomConfig(name="full", version="1.0.0", description="load both segmentation and depth (for all tar files, 160GB)"),
        CustomConfig(name="sample", version="1.0.0", description="load both segmentation and depth (for 1 tar file, 870MB)"),
        CustomConfig(name="depth", version="1.0.0", description="only load depth (sample)"),
        CustomConfig(name="seg", version="1.0.0", description="only load segmentation (sample)"),
    ]    # Configs initialization
    BUILDER_CONFIG_CLASS = CustomConfig
    def _info(self):
        return DatasetInfo(
            features=Features({
                "left_rgb": datasets.Image(),
                "right_rgb": datasets.Image(),
                "left_seg": datasets.Image(),
                "left_depth": datasets.Image(),
                "right_depth": datasets.Image(),
            })
        )
    def _h5_loader(self, bytes_stream, type_dataset):
        # Reference: https://github.com/dwofk/fast-depth/blob/master/dataloaders/dataloader.py#L8-L13
        f = io.BytesIO(bytes_stream)
        h5f = h5py.File(f, "r")
        left_rgb = self._read_jpg(h5f['rgb_left'][:])
        if type_dataset == 'depth':
            right_rgb = self._read_jpg(h5f['rgb_right'][:])
            left_depth = h5f['depth_left'][:].astype(np.float32)
            right_depth = h5f['depth_right'][:].astype(np.float32)
            return left_rgb, right_rgb, np.zeros((1,1)), left_depth, right_depth
        elif type_dataset == 'seg':
            left_seg = h5f['seg_left'][:]
            return left_rgb, np.zeros((1,1)), left_seg, np.zeros((1,1)), np.zeros((1,1))
        else:
            right_rgb = self._read_jpg(h5f['rgb_right'][:])
            left_seg = h5f['seg_left'][:]
            left_depth = h5f['depth_left'][:].astype(np.float32)
            right_depth = h5f['depth_right'][:].astype(np.float32)
            return left_rgb, right_rgb, left_seg, left_depth, right_depth
    def _read_jpg(self, bytes_stream):
        return Image.open(io.BytesIO(bytes_stream))
    
    def _split_generators(self, dl_manager):
        if 'full' == self.config.dataset_type:
            archives = dl_manager.download({"train":self._get_dataset_filenames(),
                                            "val":self._get_dataset_filenames()})
        else:
            archives = dl_manager.download({"train":[self._get_dataset_filenames()[0]],
                                            "val":[self._get_dataset_filenames()[0]]})
        split_metadata = dl_manager.download(_metadata_urls)
        return [
            SplitGenerator(
                name=Split.TRAIN,
                gen_kwargs={
                    "archives": [dl_manager.iter_archive(archive) for archive in archives["train"]],
                    "split_txt": split_metadata["train"]
                },
            ),
            SplitGenerator(
                name=Split.VALIDATION,
                gen_kwargs={
                    "archives": [dl_manager.iter_archive(archive) for archive in archives["val"]],
                    "split_txt": split_metadata["val"]
                },
            ),
        ]

    def _generate_examples(self, archives, split_txt):
        #print(split_txt, archives)
        with open(split_txt, encoding="utf-8") as split_f:
            all_splits = split_f.read().split('\n')
            #print(len(all_splits))
        for archive in archives:
            #print(archive)
            for path, file in archive:
                if path.split('/')[-1][:-3] not in all_splits:
                    #print(path.split('/')[-1][:-3], all_splits[0])
                    continue
                #print("added")
                left_rgb, right_rgb, left_seg, left_depth, right_depth = self._h5_loader(file.read(), self.config.dataset_type)
                yield path, {
                    "left_rgb": left_rgb,
                    "right_rgb": right_rgb,
                    "left_seg": left_seg,
                    "left_depth": left_depth,
                    "right_depth": right_depth,
                }
    def _get_dataset_filenames(self):
        fs = HfFileSystem()
        all_files = fs.ls("datasets/xingjianli/tomatotest/data")
        filenames = sorted(['/'.join(f['name'].split('/')[-2:]) for f in all_files])
        return filenames