ssl4eo_s_dataset.py

import os
import gzip
import json
import numpy as np
import rasterio
import re
from torch.utils.data import Dataset, DataLoader
import torch
#from cvtorchvision import cvtransforms
from kornia.augmentation import AugmentationSequential
import kornia
import argparse


class SSL4EO_S(Dataset):
    def __init__(self, fnames_path, root_dir, modality=['s1_grd', 's2_toa', 's3_olci', 's5p_co', 's5p_no2', 's5p_so2', 's5p_o3', 'dem'], transform_s1=None, transform_s2=None, transform_s3=None, transform_s5p=None, transform_dem=None):
        with gzip.open(fnames_path, 'rt', encoding='utf-8') as gz_file:
            self.fnames_json = json.load(gz_file)
        self.grid_ids = list(self.fnames_json.keys())
        self.root_dir = root_dir
        self.transform_s1 = transform_s1
        self.transform_s2 = transform_s2
        self.transform_s3 = transform_s3
        self.transform_s5p = transform_s5p
        self.transform_dem = transform_dem
        self.modality = modality

    def __len__(self):
        return len(self.grid_ids)

    def get_s1_s2(self,grid_id,modality):
        arrays = []
        meta_data = []
        local_grids = list(self.fnames_json[grid_id][modality].keys())
        grid_id_coord = self.fnames_json[grid_id]['grid_id_coord']
        for local_grid in local_grids:
            local_fpaths = self.fnames_json[grid_id][modality][local_grid]
            imgs = []
            meta = []
            for local_fpath in local_fpaths:
                with rasterio.open(os.path.join(self.root_dir, local_fpath)) as src:
                    img = src.read()
                if modality=='s1_grd' and self.transform_s1:
                    #img = self.transform_s1(np.transpose(img, (1, 2, 0)))
                    img = torch.from_numpy(img).unsqueeze(0)
                    img = self.transform_s1(img).squeeze(0)
                elif modality=='s2_toa' and self.transform_s2:
                    #img = self.transform_s2(np.transpose(img.astype(np.int16), (1, 2, 0)))
                    img = torch.from_numpy(img.astype(np.int16)).unsqueeze(0)
                    img = self.transform_s2(img.to(torch.float16)).squeeze(0)
                imgs.append(img)
                fname = local_fpath.split('/')[-1]
                date = re.search(r'(\d{8})T', fname).group(1)
                meta_info = f"{grid_id_coord}/{local_grid}/{date}"
                meta.append(meta_info)
            arrays.append(imgs)
            meta_data.append(meta)
        return arrays, meta_data

    def get_s3(self,grid_id):
        arrays = []
        meta_data = []
        fpaths = self.fnames_json[grid_id]['s3_olci']
        grid_id_coord = self.fnames_json[grid_id]['grid_id_coord']
        for fpath in fpaths:
            with rasterio.open(os.path.join(self.root_dir, fpath)) as src:
                img = src.read()
            if self.transform_s3:
                #img = self.transform_s3(np.transpose(img, (1, 2, 0)))
                img = torch.from_numpy(img).unsqueeze(0)
                img = self.transform_s3(img).squeeze(0)
            arrays.append(img)
            fname = fpath.split('/')[-1]
            date = re.search(r'(\d{8})T', fname).group(1)
            meta_info = f"{grid_id_coord}/{date}"
            meta_data.append(meta_info)
        return arrays, meta_data

    def get_s5p(self,grid_id,modality):
        arrays = []
        meta_data = []
        fpaths = self.fnames_json[grid_id][modality]
        grid_id_coord = self.fnames_json[grid_id]['grid_id_coord']
        for fpath in fpaths:
            with rasterio.open(os.path.join(self.root_dir, fpath)) as src:
                img = src.read()
            if self.transform_s5p:
                #img = self.transform_s5p(np.transpose(img, (1, 2, 0)))
                img = torch.from_numpy(img).unsqueeze(0)
                img = self.transform_s5p(img).squeeze(0)
            arrays.append(img)
            fname = fpath.split('/')[-1]
            match = re.search(r'(\d{4})-(\d{2})-(\d{2})', fname)
            date = f"{match.group(1)}{match.group(2)}{match.group(3)}"
            meta_info = f"{grid_id_coord}/{date}"
            meta_data.append(meta_info)
        return arrays, meta_data

    def get_dem(self,grid_id):
        fpath = self.fnames_json[grid_id]['dem'][0]
        with rasterio.open(os.path.join(self.root_dir, fpath)) as src:
            img = src.read()
        if self.transform_dem:
            #img = self.transform_dem(np.transpose(img, (1, 2, 0)))
            img = torch.from_numpy(img).unsqueeze(0)
            img = self.transform_dem(img).squeeze(0)
        return img

    def __getitem__(self, idx):
        grid_id = self.grid_ids[idx]
        grid_id_coord = self.fnames_json[grid_id]['grid_id_coord']
        sample = {}
        meta_data = {}
        # s1
        if 's1_grd' in self.modality:
            arr_s1, meta_s1 = self.get_s1_s2(grid_id,'s1_grd')
            sample['s1_grd'] = arr_s1
            meta_data['s1_grd'] = meta_s1
        # s2
        if 's2_toa' in self.modality:
            arr_s2, meta_s2 = self.get_s1_s2(grid_id,'s2_toa')
            sample['s2_toa'] = arr_s2
            meta_data['s2_toa'] = meta_s2
        # s3
        if 's3_olci' in self.modality:
            arr_s3, meta_s3 = self.get_s3(grid_id)
            sample['s3_olci'] = arr_s3
            meta_data['s3_olci'] = meta_s3
        # s5p_co
        if 's5p_co' in self.modality:
            arr_s5p_co, meta_s5p_co = self.get_s5p(grid_id,'s5p_co')
            sample['s5p_co'] = arr_s5p_co
            meta_data['s5p_co'] = meta_s5p_co
        # s5p_no2
        if 's5p_no2' in self.modality:
            arr_s5p_no2, meta_s5p_no2 = self.get_s5p(grid_id,'s5p_no2')
            sample['s5p_no2'] = arr_s5p_no2
            meta_data['s5p_no2'] = meta_s5p_no2
        # s5p_o3
        if 's5p_o3' in self.modality:
            arr_s5p_o3, meta_s5p_o3 = self.get_s5p(grid_id,'s5p_o3')
            sample['s5p_o3'] = arr_s5p_o3
            meta_data['s5p_o3'] = meta_s5p_o3
        # s5p_so2
        if 's5p_so2' in self.modality:
            arr_s5p_so2, meta_s5p_so2 = self.get_s5p(grid_id,'s5p_so2')
            sample['s5p_so2'] = arr_s5p_so2
            meta_data['s5p_so2'] = meta_s5p_so2
        # dem
        if 'dem' in self.modality:
            arr_dem = self.get_dem(grid_id)
            sample['dem'] = arr_dem
            meta_data['dem'] = grid_id_coord

        return sample, meta_data


if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('--fnames_path', type=str, default='data_loading/fnames.json.gz')
    parser.add_argument('--root_dir', type=str, default='data_loading/data')
    args = parser.parse_args()

    # transform_s1 = cvtransforms.Compose([
    #     cvtransforms.CenterCrop(224),
    #     cvtransforms.ToTensor()
    # ])
    # transform_s2 = cvtransforms.Compose([
    #     cvtransforms.CenterCrop(224),
    #     cvtransforms.ToTensor()
    # ])
    # transform_s3 = cvtransforms.Compose([
    #     cvtransforms.CenterCrop(96),
    #     cvtransforms.ToTensor()
    # ])
    # transform_s5p = cvtransforms.Compose([
    #     cvtransforms.CenterCrop(28),
    #     cvtransforms.ToTensor()
    # ])
    # transform_dem = cvtransforms.Compose([
    #     cvtransforms.CenterCrop(960),
    #     cvtransforms.ToTensor()
    # ])
    transform_s1 = AugmentationSequential(
        #kornia.augmentation.SmallestMaxSize(264),
        kornia.augmentation.CenterCrop(224),
    )
    transform_s2 = AugmentationSequential(
        #kornia.augmentation.SmallestMaxSize(264),
        kornia.augmentation.CenterCrop(224),
    )
    transform_s3 = AugmentationSequential(
        kornia.augmentation.SmallestMaxSize(96),
        kornia.augmentation.CenterCrop(96),
    )
    transform_s5p = AugmentationSequential(
        kornia.augmentation.SmallestMaxSize(28),
        kornia.augmentation.CenterCrop(28),
    )
    transform_dem = AugmentationSequential(
        kornia.augmentation.SmallestMaxSize(960),
        kornia.augmentation.CenterCrop(960),
    )


    ssl4eo_s = SSL4EO_S(args.fnames_path, args.root_dir, transform_s1=transform_s1, transform_s2=transform_s2, transform_s3=transform_s3, transform_s5p=transform_s5p, transform_dem=transform_dem)
    dataloader = DataLoader(ssl4eo_s, batch_size=1, shuffle=True, num_workers=0) # batch size can only be 1 because of varying number of images per grid

    for i, (sample, meta_data) in enumerate(dataloader):
        #print(i)
        print('Grid ID:', meta_data['dem'][0])
        print(sample.keys())
        print(meta_data.keys())

        
        print('### S1 GRD ###')
        print('Number of s1 local patches:', len(meta_data['s1_grd']), '  ', 'Number of time stamps for first local patch:', len(meta_data['s1_grd'][0]))
        print('Example for one image:', sample['s1_grd'][0][0].shape, meta_data['s1_grd'][0][0])
        print('### S2 TOA ###')
        print('Number of s2 local patches:', len(meta_data['s2_toa']), '  ', 'Number of time stamps for first local patch:', len(meta_data['s2_toa'][0]))
        print('Example for one image:', sample['s2_toa'][0][0].shape, meta_data['s2_toa'][0][0])
        print('### S3 OLCI ###')
        print('Number of s3 time stamps:', len(meta_data['s3_olci']))
        print('Example for one image:', sample['s3_olci'][0].shape, meta_data['s3_olci'][0])
        print('### S5P ###')
        print('Number of s5p time stamps for CO/NO2/O3/SO2:', len(meta_data['s5p_co']), len(meta_data['s5p_no2']), len(meta_data['s5p_o3']), len(meta_data['s5p_so2']))
        print('Example for one CO image:', sample['s5p_co'][0].shape, meta_data['s5p_co'][0])
        print('Example for one NO2 image:', sample['s5p_no2'][0].shape, meta_data['s5p_no2'][0])
        print('Example for one O3 image:', sample['s5p_o3'][0].shape, meta_data['s5p_o3'][0])
        print('Example for one SO2 image:', sample['s5p_so2'][0].shape, meta_data['s5p_so2'][0])
        print('### DEM ###')
        print('One DEM image for the grid:', sample['dem'].shape, meta_data['dem'][0])

        break