File size: 26,724 Bytes
068792f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 | {
"cells": [
{
"cell_type": "markdown",
"id": "f541ffd4",
"metadata": {},
"source": [
"# Synthetic High-Resolution DEM Generation for Marrakech, Morocco\n",
"# Using Only McKinley Dataset for Training\n",
"\n",
"This notebook implements the full pipeline, training only on the McKinley dataset to generate a model for super-resolving 30m SRTM to 10m DEMs fused with Sentinel-2 imagery for Marrakech, Morocco.\n",
"\n",
"**Key Assumptions:**\n",
"- Training on McKinley Mine NM high-res LiDAR DEM.\n",
"- Inference on Marrakech mountain area.\n",
"- Adapted DeepDEM model with 7 input channels.\n",
"\n",
"Run cells sequentially."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b7aa9465",
"metadata": {},
"outputs": [],
"source": [
"# Cell 1: Install Dependencies\n",
"!pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118\n",
"!pip install pytorch-lightning torchgeo segmentation-models-pytorch rasterio geopandas albumentations scipy gdown earthengine-api\n",
"!apt-get install -y libspatialindex-dev libgdal-dev\n",
"!pip install gdal==$(gdal-config --version)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c4f399ac",
"metadata": {},
"outputs": [],
"source": [
"# Cell 2: Mount Google Drive and Set Up Directories\n",
"from google.colab import drive\n",
"drive.mount('/content/drive')\n",
"%cd /content/drive/MyDrive/DEM_Project\n",
"!mkdir -p Training_Data/McKinley Inference_Data/Marrackech Models"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "ed508a36",
"metadata": {},
"outputs": [],
"source": [
"# Cell 3: Clone DeepDEM Repo and Adapt for Our Use Case\n",
"!git clone https://github.com/uw-cryo/DeepDEM.git\n",
"%cd DeepDEM\n",
"\n",
"# Adapt model for our inputs: Modify task_module.py to accept ['dsm', 'ortho_r', 'ortho_g', 'ortho_b', 'ortho_nir', 'ndvi', 'nodata_mask'] (7 channels)\n",
"# Set model in_channels=7, out_channels=1 (residuals)\n",
"# For simplicity, assume manual edit or duplicate code here.\n",
"\n",
"import os\n",
"os.environ['PYTHONPATH'] += ':/content/drive/MyDrive/DEM_Project/DeepDEM'"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "d7bb1f40",
"metadata": {},
"outputs": [],
"source": [
"# Cell 4: Authenticate and Initialize Earth Engine\n",
"from google.colab import auth\n",
"import ee\n",
"\n",
"# 1. Authenticate your Google user\n",
"auth.authenticate_user()\n",
"\n",
"# 2. Initialize Earth Engine with your Google Cloud Project ID\n",
"# REPLACE 'your-gcp-project-id' with the actual ID of your project\n",
"try:\n",
" ee.Initialize(project='dem-collab')\n",
" print(\"Earth Engine initialized successfully!\")\n",
"except ee.EEException as e:\n",
" print(f\"Error during initialization: {e}\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "091b3f03",
"metadata": {},
"outputs": [],
"source": [
"# Cell 5: Data Acquisition Function (SRTM + Sentinel-2 for McKinley and Marrakech)\n",
"import os\n",
"\n",
"def fetch_gee_data(bbox, output_dir, dataset_name):\n",
" os.makedirs(output_dir, exist_ok=True)\n",
"\n",
" geom = ee.Geometry.BBox(*bbox) # Define geometry for region\n",
"\n",
" # SRTM (30m)\n",
" srtm = ee.Image('CGIAR/SRTM90_V4').clip(geom).rename('dsm')\n",
" task_srtm = ee.batch.Export.image.toDrive(\n",
" image=srtm,\n",
" description=f'{dataset_name}_srtm',\n",
" folder=output_dir.split('/')[-1],\n",
" scale=30,\n",
" fileFormat='GeoTIFF',\n",
" region=geom\n",
" )\n",
" task_srtm.start()\n",
"\n",
" # Sentinel-2 (10m, cloud-free median, RGB + NIR)\n",
" # Sharper image: sort by cloud cover and take the best one from a good season\n",
" collection = ee.ImageCollection('COPERNICUS/S2_SR_HARMONIZED') \\\n",
" .filterBounds(geom) \\\n",
" .filterDate('2023-06-01', '2023-10-31') \\\n",
" .filter(ee.Filter.lt('CLOUDY_PIXEL_PERCENTAGE', 10)) \\\n",
" .sort('CLOUDY_PIXEL_PERCENTAGE')\n",
"\n",
" num_images = collection.size().getInfo()\n",
" print(f'For {dataset_name} bbox: {num_images} Sentinel-2 images found.')\n",
"\n",
" # Get the best image\n",
" best_image = collection.first()\n",
"\n",
" # Export the raw 4-band image for the model\n",
" sentinel_raw = best_image.select(['B4','B3','B2','B8'])\n",
" task_s2_raw = ee.batch.Export.image.toDrive(\n",
" image=sentinel_raw,\n",
" description=f'{dataset_name}_sentinel', # Keep original name for downstream tasks\n",
" folder=output_dir.split('/')[-1],\n",
" scale=10,\n",
" fileFormat='GeoTIFF',\n",
" region=geom\n",
" )\n",
" task_s2_raw.start()\n",
"\n",
" # Export a separate, color-corrected visual version for inspection\n",
" sentinel_viz = best_image.visualize(min=0, max=3000, bands=['B4', 'B3', 'B2'])\n",
" task_s2_viz = ee.batch.Export.image.toDrive(\n",
" image=sentinel_viz,\n",
" description=f'{dataset_name}_sentinel_viz',\n",
" folder=output_dir.split('/')[-1],\n",
" scale=10,\n",
" fileFormat='GeoTIFF',\n",
" region=geom\n",
" )\n",
" task_s2_viz.start()\n",
"\n",
"\n",
"# Bounding boxes\n",
"bbox_mckinley = [-109.03892074228675, 35.58282920746211, -108.87077846472735, 35.736434167381475]\n",
"fetch_gee_data(bbox_mckinley, '/content/drive/MyDrive/DEM_Project/Training_Data/McKinley', 'mckinley')\n",
"\n",
"bbox_marrakech = [-8.1, 31.5, -7.9, 31.7]\n",
"fetch_gee_data(bbox_marrakech, '/content/drive/MyDrive/DEM_Project/Inference_Data/Marrakech', 'marrakech')"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "5f9b0934",
"metadata": {},
"outputs": [],
"source": [
"# Cell 6: Download High-Resolution DEM Tiles and Merge (Only for McKinley)\n",
"!pip install boto3 gdal retry\n",
"!apt install gdal-bin\n",
"import boto3\n",
"import os\n",
"import shutil\n",
"from botocore import UNSIGNED\n",
"from botocore.client import Config\n",
"from retry import retry\n",
"\n",
"# S3 Configuration\n",
"endpoint_url = 'https://opentopography.s3.sdsc.edu'\n",
"client = boto3.client('s3', endpoint_url=endpoint_url, config=Config(signature_version=UNSIGNED))\n",
"\n",
"# Temp and data dirs\n",
"temp_base = '/content/hr_temp'\n",
"data_base = '/content/drive/MyDrive/DEM_Project/Training_Data'\n",
"os.makedirs(temp_base, exist_ok=True)\n",
"os.makedirs(data_base, exist_ok=True)\n",
"\n",
"# Only McKinley\n",
"datasets = {'mckinley': 'NM23_McKinley'}\n",
"folder_names = {'mckinley': 'McKinley'}\n",
"\n",
"@retry(tries=3, delay=2, backoff=2)\n",
"def download_file_with_retry(bucket, key, filename):\n",
" client.download_file(Bucket=bucket, Key=key, Filename=filename)\n",
"\n",
"!df -h /content\n",
"for local_name, s3_dir in datasets.items():\n",
" temp_dir = os.path.join(temp_base, local_name)\n",
" dataset_dir = os.path.join(data_base, folder_names[local_name])\n",
" os.makedirs(temp_dir, exist_ok=True)\n",
" os.makedirs(dataset_dir, exist_ok=True)\n",
"\n",
" output_tif = os.path.join(dataset_dir, f'{local_name}_hr_dem.tif')\n",
" if os.path.exists(output_tif):\n",
" print(f'Merged DEM already exists for {local_name}: {output_tif}, skipping download and merge.')\n",
" continue\n",
"\n",
" paginator = client.get_paginator('list_objects_v2')\n",
" prefix = f'{s3_dir}/{s3_dir}_be/'\n",
" downloaded_files = []\n",
" try:\n",
" for page in paginator.paginate(Bucket='raster', Prefix=prefix):\n",
" for obj in page.get('Contents', []):\n",
" key = obj['Key']\n",
" if key.endswith('.tif'):\n",
" file_path = os.path.join(temp_dir, os.path.basename(key))\n",
" # Check if the tile file already exists\n",
" if os.path.exists(file_path):\n",
" print(f'Tile {os.path.basename(key)} already exists for {local_name}, skipping download.')\n",
" downloaded_files.append(file_path)\n",
" continue\n",
" try:\n",
" download_file_with_retry('raster', key, file_path)\n",
" downloaded_files.append(file_path)\n",
" print(f'Downloaded {os.path.basename(key)} for {local_name}')\n",
" except Exception as e:\n",
" print(f'Error downloading {key} for {local_name}: {e}')\n",
" except Exception as e:\n",
" print(f'Error listing tiles for {local_name}: {e}')\n",
"\n",
" if not downloaded_files:\n",
" print(f'No tiles downloaded for {local_name}; skipping merge.')\n",
" continue\n",
"\n",
" try:\n",
" # Using gdalbuildvrt and gdal_translate for better performance\n",
" !gdalbuildvrt merged.vrt {\" \".join(downloaded_files)}\n",
" !gdal_translate -of GTiff merged.vrt \"{output_tif}\" -co TILED=YES -co COMPRESS=DEFLATE -co NUM_THREADS=ALL_CPUS\n",
" print(f'Merged tiles to TIFF for {local_name}: {output_tif}')\n",
" except Exception as e:\n",
" print(f'Error merging tiles for {local_name}: {e}')\n",
" continue\n",
" # Removed the shutil.rmtree(temp_dir) line as requested\n",
"\n",
"!df -h /content\n",
"print('Download complete for McKinley!')"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "7f82f16c",
"metadata": {},
"outputs": [],
"source": [
"# Cell 7: Data Preprocessing (Only for McKinley and Marrakech)\n",
"import rasterio\n",
"from rasterio.enums import Resampling\n",
"import numpy as np\n",
"from scipy.ndimage import gaussian_filter\n",
"import os\n",
"\n",
"folder_names = {'mckinley': 'McKinley', 'marrakech': 'Marrakech'}\n",
"\n",
"def preprocess_dataset(dataset_name, is_training=True, custom_base=None):\n",
" if custom_base:\n",
" base_dir = custom_base\n",
" else:\n",
" base_dir = f'/content/drive/MyDrive/DEM_Project/Training_Data/{folder_names[dataset_name]}'\n",
" \n",
" srtm_path = os.path.join(base_dir, f'{dataset_name}_srtm.tif')\n",
" s2_path = os.path.join(base_dir, f'{dataset_name}_sentinel.tif')\n",
" hr_path = os.path.join(base_dir, f'{dataset_name}_hr_dem.tif') if is_training else None\n",
" output_dir = base_dir\n",
"\n",
" with rasterio.open(srtm_path) as srtm_src, rasterio.open(s2_path) as s2_src:\n",
" target_shape = (s2_src.height, s2_src.width)\n",
" srtm = srtm_src.read(1, out_shape=target_shape, resampling=Resampling.cubic)\n",
"\n",
" s2 = s2_src.read()\n",
" r, g, b, nir = s2\n",
"\n",
" ndvi = (nir - r) / (nir + r + 1e-10)\n",
"\n",
" mask = np.where(srtm == srtm_src.nodata, 1, 0).astype(np.float32)\n",
"\n",
" if is_training:\n",
" with rasterio.open(hr_path) as hr_src:\n",
" hr = hr_src.read(1, out_shape=target_shape, resampling=Resampling.cubic) if hr_src.shape != target_shape else hr_src.read(1)\n",
"\n",
" trend = gaussian_filter(hr, sigma=5)\n",
" residual = hr - trend\n",
"\n",
" target_profile = s2_src.profile\n",
" target_profile['count'] = 1\n",
" with rasterio.open(os.path.join(output_dir, 'target.tif'), 'w', **target_profile) as dst:\n",
" dst.write(residual, 1)\n",
"\n",
" input_profile = s2_src.profile\n",
" input_profile['count'] = 7\n",
" with rasterio.open(os.path.join(output_dir, 'input.tif'), 'w', **input_profile) as dst:\n",
" dst.write(srtm, 1)\n",
" dst.write(r, 2)\n",
" dst.write(g, 3)\n",
" dst.write(b, 4)\n",
" dst.write(nir, 5)\n",
" dst.write(ndvi, 6)\n",
" dst.write(mask, 7)\n",
"\n",
"# Preprocess McKinley\n",
"preprocess_dataset('mckinley', is_training=True)\n",
"\n",
"# Preprocess Marrakech (no HR)\n",
"preprocess_dataset('marrakech', is_training=False, custom_base='/content/drive/MyDrive/DEM_Project/Inference_Data/Marrakech')\n",
"\n",
"!df -h /content/drive/MyDrive"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "7933d058",
"metadata": {},
"outputs": [],
"source": [
"# Cell 8: Custom Dataset Class\n",
"import albumentations as A\n",
"from albumentations.pytorch import ToTensorV2\n",
"from torch.utils.data import Dataset\n",
"import rasterio.windows\n",
"\n",
"class CustomDEMDataset(Dataset):\n",
" def __init__(self, data_dirs, tile_size=256, transform=None):\n",
" self.pairs = []\n",
" for d_dir in data_dirs:\n",
" input_path = os.path.join(d_dir, 'input.tif')\n",
" target_path = os.path.join(d_dir, 'target.tif')\n",
" if os.path.exists(input_path) and os.path.exists(target_path):\n",
" self.pairs.append((input_path, target_path))\n",
" self.tile_size = tile_size\n",
" self.transform = transform or A.Compose([\n",
" A.RandomCrop(height=tile_size, width=tile_size),\n",
" A.RandomRotate90(),\n",
" A.HorizontalFlip(),\n",
" A.VerticalFlip(),\n",
" A.GaussNoise(var_limit=(0.01, 0.01)),\n",
" ToTensorV2()\n",
" ])\n",
"\n",
" def __len__(self):\n",
" return len(self.pairs) * 50\n",
"\n",
" def __getitem__(self, idx):\n",
" input_path, target_path = self.pairs[idx % len(self.pairs)]\n",
" with rasterio.open(input_path) as inp, rasterio.open(target_path) as tgt:\n",
" max_col = inp.width - self.tile_size\n",
" max_row = inp.height - self.tile_size\n",
" col_off = np.random.randint(0, max_col + 1)\n",
" row_off = np.random.randint(0, max_row + 1)\n",
" window = rasterio.windows.Window(col_off, row_off, self.tile_size, self.tile_size)\n",
" input_data = inp.read(window=window)\n",
" target_data = tgt.read(1, window=window)\n",
"\n",
" data = {'image': input_data.transpose(1,2,0).astype(np.float32), 'target': target_data.astype(np.float32)}\n",
" augmented = self.transform(image=data['image'], mask=data['target'])\n",
" return augmented['image'], augmented['mask'].unsqueeze(0)\n",
"\n",
"train_dirs = ['/content/drive/MyDrive/DEM_Project/Training_Data/McKinley']\n",
"dataset = CustomDEMDataset(train_dirs)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "46cf2c6d",
"metadata": {},
"outputs": [],
"source": [
"# Cell 9: Model and Training (Using Only McKinley)\n",
"import pytorch_lightning as pl\n",
"import segmentation_models_pytorch as smp\n",
"from torch.utils.data import DataLoader\n",
"import torch\n",
"import torch.nn as nn\n",
"\n",
"class DeepDEMRefinement(pl.LightningModule):\n",
" def __init__(self, lr=1e-4):\n",
" super().__init__()\n",
" self.model = smp.Unet(encoder_name='resnet34', in_channels=7, classes=1, activation=None)\n",
" self.loss_fn = nn.L1Loss()\n",
" self.lr = lr\n",
"\n",
" def forward(self, x):\n",
" return self.model(x)\n",
"\n",
" def training_step(self, batch, batch_idx):\n",
" inputs, targets = batch\n",
" preds = self(inputs)\n",
" loss = self.loss_fn(preds, targets)\n",
" self.log('train_loss', loss)\n",
" return loss\n",
"\n",
" def configure_optimizers(self):\n",
" return torch.optim.Adam(self.parameters(), lr=self.lr)\n",
"\n",
"class DEMDataModule(pl.LightningDataModule):\n",
" def __init__(self, train_dirs, batch_size=4):\n",
" super().__init__()\n",
" self.train_dataset = CustomDEMDataset(train_dirs)\n",
" self.batch_size = batch_size\n",
"\n",
" def train_dataloader(self):\n",
" return DataLoader(self.train_dataset, batch_size=self.batch_size, shuffle=True, num_workers=2)\n",
"\n",
"model = DeepDEMRefinement()\n",
"datamodule = DEMDataModule(train_dirs)\n",
"trainer = pl.Trainer(max_epochs=5, accelerator='gpu', devices=1) # Training for 5 epochs\n",
"trainer.fit(model, datamodule)\n",
"\n",
"trainer.save_checkpoint('/content/drive/MyDrive/DEM_Project/Models/deepdem_model.ckpt')"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "30872060",
"metadata": {},
"outputs": [],
"source": [
"# Cell 10: Inference for Marrakech\n",
"model = DeepDEMRefinement.load_from_checkpoint('/content/drive/MyDrive/DEM_Project/Models/deepdem_model.ckpt')\n",
"model.eval()\n",
"model.to('cuda' if torch.cuda.is_available() else 'cpu')\n",
"\n",
"input_path = '/content/drive/MyDrive/DEM_Project/Inference_Data/Marrakech/input.tif'\n",
"with rasterio.open(input_path) as src:\n",
" input_data = src.read().astype(np.float32)\n",
" trend = gaussian_filter(input_data[0], sigma=5)\n",
"\n",
" input_tensor = torch.from_numpy(input_data).unsqueeze(0).to(model.device)\n",
"\n",
" with torch.no_grad():\n",
" residual_pred = model(input_tensor)\n",
"\n",
" synth_dem = residual_pred.squeeze().cpu().numpy() + trend\n",
"\n",
" profile = src.profile\n",
" profile['count'] = 1\n",
" with rasterio.open('/content/drive/MyDrive/DEM_Project/synth_dem_marrakech.tif', 'w', **profile) as dst:\n",
" dst.write(synth_dem, 1)\n",
"\n",
"print('Synthetic DEM generated for Marrakech!')"
]
},
{
"cell_type": "markdown",
"id": "8943e470",
"metadata": {},
"source": [
"# Quick correctness checks\n",
"\n",
"This section runs a few sanity checks on the trained model and data:\n",
"\n",
"- Validate shapes, CRS, and basic channel statistics of `input.tif` and `target.tif`\n",
"- Compute masked MAE/RMSE on random training crops (McKinley) to gauge training fit\n",
"- Flag obvious issues (e.g., all-zeros bands, nodata dominance)\n",
"\n",
"Run cells in order after training has completed."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "3c35167d",
"metadata": {},
"outputs": [],
"source": [
"# Check 1: Inspect input/target rasters (McKinley)\n",
"import rasterio\n",
"import numpy as np\n",
"from pathlib import Path\n",
"\n",
"train_dir = Path('/content/drive/MyDrive/DEM_Project/Training_Data/McKinley')\n",
"input_path = train_dir / 'input.tif'\n",
"target_path = train_dir / 'target.tif'\n",
"\n",
"issues = []\n",
"\n",
"with rasterio.open(input_path) as src:\n",
" print('INPUT:')\n",
" print({'count': src.count, 'width': src.width, 'height': src.height, 'crs': str(src.crs), 'dtype': src.dtypes})\n",
" data = src.read(out_dtype='float32')\n",
" nodata = src.nodata\n",
" band_stats = []\n",
" for i in range(src.count):\n",
" b = data[i]\n",
" if nodata is not None:\n",
" mask = b == nodata\n",
" valid = np.where(mask, np.nan, b)\n",
" else:\n",
" valid = b\n",
" mask = np.zeros_like(b, dtype=bool)\n",
" s = {\n",
" 'band': i+1,\n",
" 'nan_frac': float(np.mean(np.isnan(valid))),\n",
" 'nodata_frac': float(np.mean(mask)),\n",
" 'min': float(np.nanmin(valid)),\n",
" 'max': float(np.nanmax(valid)),\n",
" 'mean': float(np.nanmean(valid)),\n",
" 'std': float(np.nanstd(valid)),\n",
" }\n",
" band_stats.append(s)\n",
" print('Input band stats (1:dsm, 2:R, 3:G, 4:B, 5:NIR, 6:NDVI, 7:mask):')\n",
" for s in band_stats:\n",
" print(s)\n",
" # Basic checks\n",
" if band_stats[5]['min'] < -1.01 or band_stats[5]['max'] > 1.01:\n",
" issues.append('NDVI out of expected [-1,1] range; check scaling and bands (R,NIR indices).')\n",
" if band_stats[6]['mean'] < 0.01 and band_stats[6]['max'] < 0.5:\n",
" issues.append('Mask band appears mostly zeros; ensure mask=1 at nodata pixels, 0 elsewhere.')\n",
"\n",
"with rasterio.open(target_path) as src:\n",
" print('\\nTARGET:')\n",
" print({'count': src.count, 'width': src.width, 'height': src.height, 'crs': str(src.crs), 'dtype': src.dtypes})\n",
" t = src.read(1, out_dtype='float32')\n",
" print({'min': float(np.nanmin(t)), 'max': float(np.nanmax(t)), 'mean': float(np.nanmean(t)), 'std': float(np.nanstd(t))})\n",
" if np.allclose(t, 0):\n",
" issues.append('Target residual is all zeros; check HR DEM loading and detrending step.')\n",
"\n",
"print('\\nPotential issues:')\n",
"print(issues if issues else 'None detected')"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "ad78e542",
"metadata": {},
"outputs": [],
"source": [
"# Check 2: Compute quick masked MAE/RMSE on random training crops\n",
"import torch\n",
"import torch.nn.functional as F\n",
"from torch.utils.data import DataLoader\n",
"import numpy as np\n",
"\n",
"# Reuse dataset and model classes already defined earlier\n",
"try:\n",
" _ = CustomDEMDataset\n",
"except NameError:\n",
" raise RuntimeError('CustomDEMDataset not defined; run earlier cells first.')\n",
"\n",
"try:\n",
" _ = DeepDEMRefinement\n",
"except NameError:\n",
" raise RuntimeError('DeepDEMRefinement not defined; run training cells first.')\n",
"\n",
"# Load model\n",
"ckpt = '/content/drive/MyDrive/DEM_Project/Models/deepdem_model.ckpt'\n",
"model = DeepDEMRefinement.load_from_checkpoint(ckpt)\n",
"model.eval()\n",
"device = 'cuda' if torch.cuda.is_available() else 'cpu'\n",
"model.to(device)\n",
"\n",
"# Small eval dataset with deterministic crops\n",
"np.random.seed(42)\n",
"transform = A.Compose([\n",
" A.RandomCrop(height=256, width=256),\n",
" ToTensorV2()\n",
"])\n",
"\n",
"val_ds = CustomDEMDataset([str(train_dir)], tile_size=256, transform=transform)\n",
"val_loader = DataLoader(val_ds, batch_size=4, shuffle=False, num_workers=2)\n",
"\n",
"maes, rmses = [], []\n",
"with torch.no_grad():\n",
" for i, (x, y) in enumerate(val_loader):\n",
" if i >= 10: # ~40 tiles\n",
" break\n",
" x = x.to(device)\n",
" y = y.to(device)\n",
" pred = model(x)\n",
" # If mask channel included, optionally down-weight masked pixels\n",
" mask = x[:, 6:7] # channel 7\n",
" valid = (mask < 0.5).float()\n",
" diff = (pred - y) * valid\n",
" denom = valid.sum().clamp_min(1.0)\n",
" mae = diff.abs().sum() / denom\n",
" rmse = torch.sqrt((diff.pow(2).sum() / denom))\n",
" maes.append(mae.item())\n",
" rmses.append(rmse.item())\n",
"\n",
"print({'MAE_mean': float(np.mean(maes)), 'MAE_std': float(np.std(maes)), 'RMSE_mean': float(np.mean(rmses)), 'RMSE_std': float(np.std(rmses)), 'tiles': len(maes)*val_loader.batch_size})\n",
"\n",
"if np.mean(rmses) > 8.0:\n",
" print('Warning: High RMSE for residuals. Training may be underfit or target scaling may be off.')\n",
"else:\n",
" print('Residual error looks reasonable for the training run.')"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "def70a98",
"metadata": {},
"outputs": [],
"source": [
"# Check 3: Sanity-check inference output alignment vs input for Marrakech\n",
"from scipy.ndimage import gaussian_filter\n",
"\n",
"marrakech_input = '/content/drive/MyDrive/DEM_Project/Inference_Data/Marrakech/input.tif'\n",
"marrakech_out = '/content/drive/MyDrive/DEM_Project/synth_dem_marrakech.tif'\n",
"\n",
"with rasterio.open(marrakech_input) as src_in, rasterio.open(marrakech_out) as src_out:\n",
" print('INFERENCE INPUT:', {'shape': (src_in.count, src_in.height, src_in.width), 'crs': str(src_in.crs), 'transform': tuple(src_in.transform)})\n",
" print('SYNTH OUTPUT:', {'shape': (src_out.count, src_out.height, src_out.width), 'crs': str(src_out.crs), 'transform': tuple(src_out.transform)})\n",
" if src_in.crs != src_out.crs:\n",
" print('Warning: CRS mismatch between input and output!')\n",
" if (src_in.height != src_out.height) or (src_in.width != src_out.width):\n",
" print('Warning: Dimension mismatch between input and output!')\n",
"\n",
" out_dem = src_out.read(1).astype('float32')\n",
" # Simple terrain sanity: residual-added trend should correlate with SRTM trend\n",
" srtm = src_in.read(1).astype('float32')\n",
" trend = gaussian_filter(srtm, sigma=5)\n",
" corr = np.corrcoef(trend.flatten(), out_dem.flatten())[0,1]\n",
" print('Correlation between SRTM trend and synthetic DEM:', float(corr))\n",
" if corr < 0.5:\n",
" print('Low correlation; output may be noisy or misaligned.')"
]
}
],
"metadata": {
"language_info": {
"name": "python"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
|