scripts/crop_tiles.py

'''
Crop Dataset to 333x333 pixels for faster processing and testing.
This script reads the original DEM images, crops them to the specified size, and saves the cropped versions in a new directory.

The annotations are also updated accordingly to reflect the new image dimensions.
'''
#------------------------------------------------------------------------
#%% IMPORTS
from pathlib import Path
import math
import rasterio
from rasterio.windows import Window
from shapely.geometry import box
import geopandas as gpd
import numpy as np
from tqdm.auto import tqdm
#------------------------------------------------------------------------
#%% CONFIGURATION

AREAS = {'ATA_MV', 'BRA_SP', 'CHN_WS', 'ESP_EH', 'FIN_LM', 'GER_BN', 'IDN_SV', 'KAZ_AC', 'KSA_WA', 'NAM_HF', 'NZL_KP', 'PHL_TA', 'USA_GC'}

#AREAS = {'ATA_MV'}

WIDTH = 256
HEIGHT = 256

#------------------------------------------------------------------------
for KEY_ID in tqdm(AREAS):

    IMG_DIR = Path(f'/home/sabrina/Documents/Tese/05_Dataset/MatchGeo-DEM-v1/data/{KEY_ID}')
    CROPPED_IMG_DIR = Path(IMG_DIR, 'tiles')
    CROPPED_IMG_DIR.mkdir(exist_ok=True)
    # CROP IMAGES
    img_files = sorted(IMG_DIR.glob(f'{KEY_ID}.tif'), key=lambda x: x.stem)
    dicto_records = {
        'tile_id' : [],
        'row_idx' : [],
        'col_idx' : [],
        'geometry': []
    }

    for img_file in tqdm(img_files, desc="Cropping images"):

        # Create a folder for each City in the cropped directory
        city = img_file.parent.name
        with rasterio.open(img_file) as src:
            crs = src.crs
            img = src.read(1)

            nodata = src.nodata if src.nodata is not None else -9999

            n_rows = math.ceil(src.height / HEIGHT)
            n_cols = math.ceil(src.width / WIDTH)

            padded_w = n_cols * WIDTH
            padded_h = n_rows * HEIGHT
            pad_right = padded_w - src.width
            pad_bottom = padded_h - src.height

            if pad_right > 0 or pad_bottom > 0:
                img = np.pad(
                    img,
                    ((0, pad_bottom), (0, pad_right)),
                    mode='constant',
                    constant_values=nodata
                )

            #----------------------------------------------
            for row in range(n_rows):
                for col in range(n_cols):
                    row_off = row * HEIGHT
                    col_off = col * WIDTH

                    tile = img[row_off:row_off + HEIGHT, col_off:col_off + WIDTH]

                    # Skip completely uniform (null) tiles
                    if np.unique(tile).size == 1:
                        print("Skipping null tile")
                        continue

                    window = Window(
                        col_off=col_off,
                        row_off=row_off,
                        width=WIDTH,
                        height=HEIGHT
                    )

                    transform = src.window_transform(window)
                    cropped_image = src.read(1, window=window)
                    left, bottom, right, top = src.window_bounds(window)
                    
                    profile = src.profile.copy()
                    profile.update({
                        "height": HEIGHT,
                        "width": WIDTH,
                        "transform": transform,
                        "nodata": nodata  # ensure nodata is set correctly
                    })

                    out_file = Path(CROPPED_IMG_DIR, KEY_ID +  f'_{row+1:03d}_{col+1:03d}.tif')
                    

                    with rasterio.open(out_file, 'w', **profile) as dst:
                        dst.write(cropped_image, 1)

                    dicto_records['tile_id'].append(f"{img_file.stem}_{row+1:03d}_{col+1:03d}")
                    dicto_records['row_idx'].append(row+1)
                    dicto_records['col_idx'].append(col+1)
                    dicto_records['geometry'].append(box(left, bottom, right, top))



    print('Saving vector..')
    TILES_FILE = Path(IMG_DIR, KEY_ID + '_tiles.geojson')
    gdf = gpd.GeoDataFrame(dicto_records, crs = crs, geometry='geometry')

    gdf.to_file(TILES_FILE, driver='GEOJSON', mode='a')
    print(f"Saved files to {CROPPED_IMG_DIR}")