s2_res / verify_examples.py
frankaisabell's picture
Add MODIS-to-S2 checkpoint, examples and validation data
b0ec833 verified
Raw History Blame
4.1 kB
"""Run the portable package in an isolated environment against original predictions."""
import importlib.metadata
import json
import os
from pathlib import Path
import shutil
import subprocess
import sys
import tempfile
import time
import numpy as np
import rasterio
import torch
from affine import Affine
from modis_s2.io import ROOT, load_model, load_sample, predict, write_prediction
from modis_s2.evaluate import evaluate
torch.set_num_threads(2)
started = time.monotonic()
assert not any(("ndvi-dit/lib/python" in x and "site-packages" in x) or "/repos/modis-to-s2" in x for x in sys.path)
assert not (Path(sys.prefix) / "pyvenv.cfg").read_text().split("include-system-site-packages = ")[1].splitlines()[0].lower() == "true"
assert Path(torch.__file__).resolve().is_relative_to(Path(sys.prefix).resolve())
assert Path(np.__file__).resolve().is_relative_to(Path(sys.prefix).resolve())
model = load_model("cpu")
records = []
for directory in sorted((ROOT/"demo").iterdir()):
if not directory.is_dir():
continue
begin = time.monotonic()
inputs, metadata = load_sample(directory)
values = predict(model, inputs)
reference = np.load(directory/"reference_cpu.npy", allow_pickle=False)
np.testing.assert_allclose(values, reference, atol=1e-5, rtol=1e-5)
assert values.shape == (256, 256) and -1 <= values.min() <= values.max() <= 1
records.append({"id": directory.name, "max_absolute_difference": float(np.abs(values-reference).max()),
"seconds": time.monotonic()-begin})
print("MATCH", directory.name, records[-1], flush=True)
assert len(records) == 10
sample = ROOT/"demo"/records[0]["id"]
with tempfile.TemporaryDirectory(dir=ROOT.parent) as temporary:
temporary = Path(temporary)
inputs_only = temporary/"inputs_only"
inputs_only.mkdir()
for name in ("inputs.npz", "metadata.json"):
shutil.copyfile(sample/name, inputs_only/name)
result = temporary/"prediction.tif"
subprocess.run([sys.executable, "-m", "modis_s2.predict", "--sample", str(inputs_only),
"--output", str(result)], check=True)
_, metadata = load_sample(sample)
reference = np.load(sample/"reference_cpu.npy", allow_pickle=False)
with rasterio.open(result) as dataset:
assert dataset.count == 1 and dataset.dtypes == ("float32",)
assert dataset.crs == rasterio.crs.CRS.from_string(metadata["geometry"]["crs"])
assert dataset.transform == Affine(*metadata["geometry"]["transform"])
np.testing.assert_allclose(dataset.read(1), reference, atol=1e-5, rtol=1e-5)
with np.load(sample/"inputs.npz", allow_pickle=False) as archive:
broken = {key: archive[key] for key in archive.files if key != "s2"}
np.savez_compressed(inputs_only/"inputs.npz", **broken)
try:
load_sample(inputs_only)
except ValueError as error:
assert "missing=['s2']" in str(error)
else:
raise AssertionError("A missing sensor input was silently accepted")
metrics = evaluate(ROOT/"demo", output=ROOT/"demo_metrics_cpu.json")
result = {"passed": True, "device": "cpu", "precision": "fp32", "demo_scenes": len(records),
"prediction_tolerance": {"atol": 1e-5, "rtol": 1e-5},
"original_vs_portable_predictions": records,
"max_absolute_difference": max(x["max_absolute_difference"] for x in records),
"fresh_environment": True, "system_site_packages": False,
"training_repository_imported": False, "target_free_cli_inference": True,
"geotiff_grid_and_values_verified": True, "missing_input_rejected": True,
"demo_evaluation_completed": True,
"full_validation_predictions_recomputed": False,
"python": sys.version.split()[0],
"packages": {p: importlib.metadata.version(p) for p in
("torch", "numpy", "scipy", "safetensors", "rasterio", "huggingface-hub")},
"seconds": time.monotonic()-started}
(ROOT/"test_results.json").write_text(json.dumps(result, indent=2, allow_nan=False))
print("PORTABLE PACKAGE TESTS PASSED", flush=True)