Adds processing of images using MSX data only

thermo_scenes/scripts/preprocess_thermal_msx_only.py

@@ -0,0 +1,92 @@
+"""
+Preprocess RGB and thermal using MSX images only (it is not required to download RGB
+images one by one from the FLIR website for this script).
+
+IMPORTANT: The images cropped using this method still differ from those downloaded
+individually from the FLIR website. There is a small rotation offset, which is
+unfortunately unknown. Emperically found that the offset should be 0.165 or -0.165
+degrees for horizontal images. The best way would be to process a few images and define
+the sign.
+"""
+
+import json
+import subprocess
+from dataclasses import dataclass
+from pathlib import Path
+
+import tyro
+from PIL import Image
+from thermo_scenes.flir_thermal_images.custom_flir import CustomFlir
+
+
+@dataclass
+class Paths:
+    msx_images: Path = Path("data/datatest/")
+    """Path to the thermal data extracted from Flir One App."""
+    output_folder: Path = Path("data/output_datatest")
+    """Path to the output folder"""
+    angle: float = 0.165
+    """
+    Unknown angle to rotate the image before cropping in degrees. Empirically found that
+    good values are either 0.165 or -0.165 for horizontal images.
+    """
+
+
+def main() -> None:
+    paths = tyro.cli(Paths)
+
+    _ = CustomFlir(
+        path_to_msx_images=paths.msx_images,
+        path_to_output_folder=paths.output_folder,
+    )
+
+    # from comments here
+    # https://stackoverflow.com/questions/62095220/overlaping-raw-thermal-image-and-embedded-image-from-flir
+    # but is seems that there is an (unknown) rotation between the downloaded
+    # (one by one) and cropped frames :(
+    cropped_rgb_folder = paths.output_folder / "cropped_rgb"
+    cropped_rgb_folder.mkdir(exist_ok=True)
+    for rgb_image in (paths.output_folder / "rgb").iterdir():
+        offsets_json = subprocess.check_output(
+            [
+                "exiftool",
+                "-OffsetX",
+                "-OffsetY",
+                "-Real2IR",
+                "-EmbeddedImageWidth",
+                "-EmbeddedImageHeight",
+                "-j",
+                paths.msx_images / (rgb_image.stem + ".JPG"),
+            ]
+        )
+        offsets = json.loads(offsets_json.decode())[0]
+        real_to_ir: float = offsets["Real2IR"]
+        img_visual = Image.open(rgb_image)
+        img_visual = img_visual.rotate(paths.angle)
+
+        x_size = int(img_visual.size[0] * real_to_ir)
+        y_size = int(img_visual.size[1] * real_to_ir)
+        img_visual = img_visual.resize((x_size, y_size))
+        crop_h: int = offsets["EmbeddedImageHeight"]
+        crop_w: int = offsets["EmbeddedImageWidth"]
+
+        start_y = (img_visual.size[1] - crop_h) // 2 + int(
+            int(offsets["OffsetY"]) * real_to_ir
+        )
+        start_x = (img_visual.size[0] - crop_w) // 2 + int(
+            int(offsets["OffsetX"]) * real_to_ir
+        )
+
+        img_visual = img_visual.crop(
+            (
+                start_x,
+                start_y,
+                start_x + crop_w,
+                start_y + crop_h,
+            )
+        )
+        img_visual.save(cropped_rgb_folder / rgb_image.name)
+
+
+if __name__ == "__main__":
+    main()