SSL4EO-L-Benchmark.py

import os
import json
import shutil
import datasets
import tifffile

import pandas as pd
import numpy as np

from torchgeo.datasets.cdl import CDL
from torchgeo.datasets.nlcd import NLCD

CMAPS = {
    'nlcd': NLCD.cmap,
    'cdl': CDL.cmap,
}

S2_MEAN = [752.40087073, 884.29673756, 1144.16202635, 1297.47289228, 1624.90992062, 2194.6423161, 2422.21248945, 2581.64687018, 2368.51236873, 1805.06846033]

S2_STD = [1108.02887453, 1155.15170768, 1183.6292542, 1368.11351514, 1370.265037, 1355.55390699, 1416.51487101, 1439.3086061, 1455.52084939, 1343.48379601]

subset_names = ["etm_sr_cdl", "etm_sr_nlcd", "etm_toa_cdl", "etm_toa_nlcd", "oli_sr_cdl", "oli_sr_nlcd", "oli_tirs_toa_cdl", "oli_tirs_toa_nlcd", "etm_oli_toa_cdl", "etm_oli_toa_nlcd"]

num_classes = {
    'etm_sr_cdl': 134,
    'etm_sr_nlcd': 21,
    'etm_toa_cdl': 134,
    'etm_toa_nlcd': 21,
    'oli_sr_cdl': 134,
    'oli_sr_nlcd': 21,
    'oli_tirs_toa_cdl': 134,
    'oli_tirs_toa_nlcd': 21,
    "etm_oli_toa_nlcd": 21,
    "etm_oli_toa_cdl": 134,
}

num_channels = {
    'etm_sr_cdl': 6,
    'etm_sr_nlcd': 6,
    'etm_toa_cdl': 9,
    'etm_toa_nlcd': 9,
    'oli_sr_cdl': 7,
    'oli_sr_nlcd': 7,
    'oli_tirs_toa_cdl': 11,
    'oli_tirs_toa_nlcd': 11,
    "etm_oli_toa_nlcd": 20,
    "etm_oli_toa_cdl": 20,
}

MEAN = [0]
STD = [255]

metadata = { # TODO: check if info below is correct or not
    'etm_sr_cdl': {"bands":["B1", "B2", "B3", "B4", "B5", "B7"], "channel_wv": [485.0, 560.0, 660.0, 835.0, 1650.0, 2220.0], "mean": MEAN * 6, 'std': STD * 6}, # B6 (Thermal Band) and B8 (Panchromatic Band) are excluded
    'etm_sr_nlcd': {"bands":["B1", "B2", "B3", "B4", "B5", "B7"], "channel_wv": [485.0, 560.0, 660.0, 835.0, 1650.0, 2220.0], "mean": MEAN * 6, 'std': STD * 6},
    'etm_toa_cdl': {"bands":["B1", "B2", "B3", "B4", "B5", "B6L", "B6H", "B7", "B8"], "channel_wv": [485.0, 560.0, 660.0, 835.0, 1650.0, 10900.0, 10900.0, 2220.0, 710.0], "mean": MEAN * 9, 'std': STD * 9},
    'etm_toa_nlcd': {"bands":["B1", "B2", "B3", "B4", "B5", "B6L", "B6H", "B7", "B8"], "channel_wv": [485.0, 560.0, 660.0, 835.0, 1650.0, 10900.0, 10900.0, 2220.0, 710.0], "mean": MEAN * 9, 'std': STD * 9},
    'oli_sr_cdl': {"bands":["B1", "B2", "B3", "B4", "B5", "B6", "B7"], "channel_wv": [443.0, 482.0, 562.0, 655.0, 865.0, 1610.0, 2200.0], "mean": MEAN * 7, 'std': STD * 7},
    'oli_sr_nlcd': {"bands":["B1", "B2", "B3", "B4", "B5", "B6", "B7"], "channel_wv": [443.0, 482.0, 562.0, 655.0, 865.0, 1610.0, 2200.0], "mean": MEAN * 7, 'std': STD * 7},
    'oli_tirs_toa_cdl': {"bands":["B1", "B2", "B3", "B4", "B5", "B6", "B7", "B8", "B9", "B10", "B11"], "channel_wv": [443.0, 482.0, 562.0, 655.0, 865.0, 1610.0, 2200.0, 590.0, 1735.0, 10800.0, 12000.0], "mean": MEAN * 11, 'std': STD * 11},
    'oli_tirs_toa_nlcd': {"bands":["B1", "B2", "B3", "B4", "B5", "B6", "B7", "B8", "B9", "B10", "B11"], "channel_wv": [443.0, 482.0, 562.0, 655.0, 865.0, 1610.0, 2200.0, 590.0, 1735.0, 10800.0, 12000.0], "mean": MEAN * 11, 'std': STD * 11},
    "etm_oli_toa_nlcd": {"bands":["B1E", "B2E", "B3E", "B4E", "B5E", "B6LE", "B6HE", "B7E", "B8E", "B1O", "B2O", "B3O", "B4O", "B5O", "B6O", "B7O", "B8O", "B9O", "B10O", "B11O"], "channel_wv": [485.0, 560.0, 660.0, 835.0, 1650.0, 10900.0, 10900.0, 2220.0, 710.0, 443.0, 482.0, 562.0, 655.0, 865.0, 1610.0, 2200.0, 590.0, 1735.0, 10800.0, 12000.0], "mean": MEAN * 20, 'std': STD * 20},
    "etm_oli_toa_cdl": {"bands":["B1E", "B2E", "B3E", "B4E", "B5E", "B6LE", "B6HE", "B7E", "B8E", "B1O", "B2O", "B3O", "B4O", "B5O", "B6O", "B7O", "B8O", "B9O", "B10O", "B11O"], "channel_wv": [485.0, 560.0, 660.0, 835.0, 1650.0, 10900.0, 10900.0, 2220.0, 710.0, 443.0, 482.0, 562.0, 655.0, 865.0, 1610.0, 2200.0, 590.0, 1735.0, 10800.0, 12000.0], "mean": MEAN * 20, 'std': STD * 20},
}

class SSL4EOLBenchmarkDataset(datasets.GeneratorBasedBuilder):
    VERSION = datasets.Version("1.0.0")
    
    DATA_URL = "https://huggingface.co/datasets/GFM-Bench/SSL4EO-L-Benchmark/resolve/main/SSL4EOLBenchmark.zip" 

    SIZE = HEIGHT = WIDTH = 264

    spatial_resolution = 30

    BUILDER_CONFIGS = [
        datasets.BuilderConfig(name="default"),
        *[datasets.BuilderConfig(name=name) for name in subset_names]
    ] 

    DEFAULT_CONFIG_NAME = "oli_tirs_toa_nlcd"

    def __init__(self, *args, **kwargs):
        name = kwargs.get('config_name', None)
        self.NUM_CLASSES = num_classes[name] if name and name != "default" else num_classes['oli_tirs_toa_nlcd']
        self.NUM_CHANNELS = num_channels[name] if name and name != "default" else num_channels['oli_tirs_toa_nlcd']
        self.metadata = metadata[name] if name and name != "default" else metadata['oli_tirs_toa_nlcd']

        name = "oli_tirs_toa_nlcd" if name is None or name == "default" else name
        product = name.split('_')[-1]
        cmap = CMAPS[product]
        classes = list(cmap.keys())
        self.ordinal_map = np.zeros(max(cmap.keys()) + 1, dtype=np.int64)
        for v, k in enumerate(classes):
            self.ordinal_map[k] = v

        super().__init__(*args, **kwargs)

    def _info(self):
        metadata = self.metadata
        metadata['size'] = self.SIZE
        metadata['num_classes'] = self.NUM_CLASSES
        metadata['spatial_resolution'] = self.spatial_resolution
        return datasets.DatasetInfo(
            description=json.dumps(metadata),
            features=datasets.Features({
                "optical": datasets.Array3D(shape=(self.NUM_CHANNELS, self.HEIGHT, self.WIDTH), dtype="float32"),
                "label": datasets.Array2D(shape=(self.HEIGHT, self.WIDTH), dtype="int32"),
                "optical_channel_wv": datasets.Sequence(datasets.Value("float32")),
                "spatial_resolution": datasets.Value("int32"),
            }),
        )

    def _split_generators(self, dl_manager):
        if isinstance(self.DATA_URL, list):
            downloaded_files = dl_manager.download(self.DATA_URL)
            combined_file = os.path.join(dl_manager.download_config.cache_dir, "combined.tar.gz")        
            with open(combined_file, 'wb') as outfile:
                for part_file in downloaded_files:
                    with open(part_file, 'rb') as infile:
                        shutil.copyfileobj(infile, outfile)
            data_dir = dl_manager.extract(combined_file)
            os.remove(combined_file)
        else:
            data_dir = dl_manager.download_and_extract(self.DATA_URL)

        return [
            datasets.SplitGenerator(
                name="train",
                gen_kwargs={
                    "split": 'train',
                    "data_dir": data_dir, 
                },
            ),
            datasets.SplitGenerator(
                name="val",
                gen_kwargs={
                    "split": 'val',
                    "data_dir": data_dir,
                },
            ),
            datasets.SplitGenerator(
                name="test",
                gen_kwargs={
                    "split": 'test',
                    "data_dir": data_dir,
                },
            )
        ]

    def _generate_examples(self, split, data_dir):
        optical_channel_wv = self.metadata["channel_wv"]
        spatial_resolution = self.spatial_resolution

        data_dir = os.path.join(data_dir, "SSL4EOLBenchmark")
        if self.config.name in ["etm_oli_toa_nlcd", "etm_oli_toa_cdl"]:
            product = self.config.name.split('_')[-1] # nlcd / cdl
            metadata_oli = pd.read_csv(os.path.join(data_dir, f"metadata_oli_tirs_toa_{product}.csv"))
            metadata_etm = pd.read_csv(os.path.join(data_dir, f"metadata_etm_toa_{product}.csv"))
        
            metadata = self.sort_and_create_new_csv(metadata_oli, metadata_etm)
        else:
            metadata = pd.read_csv(os.path.join(data_dir, f"metadata_{self.config.name}.csv"))
        
        metadata = metadata[metadata["split"] == split].reset_index(drop=True)

        for index, row in metadata.iterrows():
            optical_path = os.path.join(data_dir, row.optical_path)
            optical = self._read_image(optical_path).astype(np.float32) # CxHxW

            try:
                optical_path_etm = os.path.join(data_dir, row.optical_path_etm)
                optical_etm = self._read_image(optical_path_etm).astype(np.float32)
                optical = np.concatenate((optical_etm, optical), axis=0)
            except:
                pass

            label_path = os.path.join(data_dir, row.label_path)
            label = self._read_image(label_path).astype(np.int32)
            label = self.ordinal_map[label]

            sample = {
                "optical": optical,
                "label": label,
                "optical_channel_wv": optical_channel_wv,
                "spatial_resolution": spatial_resolution,
            }

            yield f"{index}", sample
    
    def _read_image(self, image_path):
        """Read tiff image from image_path
        Args:
            image_path: 
                Image path to read from

        Return:
            image: 
                C, H, W numpy array image
        """
        image = tifffile.imread(image_path)
        if len(image.shape) == 3:
            image = np.transpose(image, (2, 0, 1))

        return image
    
    def sort_and_create_new_csv(self, metadata_oli, metadata_etm):
        def extract_number(optical_path):
            return optical_path.split('/')[1]
        
        metadata_oli['number'] = metadata_oli['optical_path'].apply(extract_number) # a number
        metadata_etm['number'] = metadata_etm['optical_path'].apply(extract_number) # a number
        
        metadata_oli = metadata_oli.sort_values('number').reset_index(drop=True)
        metadata_etm = metadata_etm.sort_values('number').reset_index(drop=True)
        
        new_rows = []
        
        i, j = 0, 0
        while i < len(metadata_oli) and j < len(metadata_etm):
            if metadata_oli.loc[i, 'number'] == metadata_etm.loc[j, 'number']:
                new_rows.append({
                    'split': metadata_oli.loc[i, 'split'],
                    'optical_path_etm': metadata_etm.loc[j, 'optical_path'],
                    'optical_path': metadata_oli.loc[i, 'optical_path'],
                    'label_path': metadata_oli.loc[i, 'label_path']
                })
                i += 1
                j += 1
            elif metadata_oli.loc[i, 'number'] < metadata_etm.loc[j, 'number']:
                i += 1
            else:
                j += 1
        
        new_df = pd.DataFrame(new_rows, columns=['split', 'optical_path_etm', 'optical_path', 'label_path'])
        return new_df