Use bundled digit templates by default

README.md

@@ -2,7 +2,7 @@
 
 Queue-position reader for Arma Reforger on Linux.
 
-The tool captures the Reforger queue screen, crops the queue-number region, and reads the orange UI digits with a small built-in matcher. It does not use Tesseract, OpenCV, or Pillow.
+The tool captures the Reforger queue screen, crops the queue-number region, and reads the orange UI digits with bundled real-game digit templates. It does not use Tesseract, OpenCV, or Pillow.
 
 ## Install
 
@@ -27,9 +27,7 @@ Build the local helpers:
 make
 ```
 
-No sudo install is required for the default Portal/PipeWire capture path.
-
-The digit matcher uses ImageMagick to render reference digits. By default it auto-selects an installed font, preferring Roboto Condensed when present and falling back to common system fonts such as DejaVu Sans.
+No sudo install is required for the default Portal/PipeWire capture path. No specific font package is required for normal use.
 
 ## Use
 
@@ -63,10 +61,10 @@ Save the captured input image while debugging:
 uv run reforger_queue_read.py --portal-window --save-input /tmp/apt-portal-window.png --show-crop --debug
 ```
 
-If font matching is weird on a specific machine, pass an installed ImageMagick font explicitly:
+The default matcher uses `templates/reforger_digits.json`. Font mode is only for debugging:
 
 ```bash
-uv run reforger_queue_read.py --portal-window --font DejaVu-Sans-Condensed --show-crop --debug
+uv run reforger_queue_read.py --portal-window --font Roboto-Condensed --show-crop --debug
 ```
 
 ## Optional KDE/KWin Mode
@@ -123,7 +121,19 @@ Use an OBS scene screenshot:
 uv run reforger_queue_read.py --obs-scene APT --show-crop --debug
 ```
 
-Validate the included dataset:
+Validate the regression set:
+
+```bash
+uv run reforger_queue_read.py --dataset datasets/regression-test-set --expect-filenames
+```
+
+Expected current result:
+
+```text
+summary: 21 passed, 0 failed
+```
+
+Validate the smaller included dataset:
 
 ```bash
 uv run reforger_queue_read.py --dataset datasets/scootz-dataset --show-crop

reforger_queue_read.py

@@ -16,19 +16,15 @@ import uuid
 
 DEFAULT_REFERENCE_SIZE = (2560, 1440)
 DEFAULT_REFERENCE_CROP = (1050, 620, 100, 60)
-DEFAULT_FONT = "auto"
+DEFAULT_TEMPLATE_SET = os.path.join(os.path.dirname(os.path.abspath(__file__)), "templates", "reforger_digits.json")
 DEFAULT_POINTSIZE = 43
 AUTO_FONT_CANDIDATES = (
     "Roboto-Condensed",
     "RobotoCondensed-Regular",
     "Roboto-Condensed-Regular",
-    "Noto-Sans-Condensed",
-    "NotoSans-Condensed",
-    "Liberation-Sans-Narrow",
-    "DejaVu-Sans-Condensed",
-    "DejaVu-Sans",
-    "Arial",
-    "Helvetica",
+    "Roboto",
+    "Adwaita-Sans",
+    "Adwaita-Sans-Bold",
 )
 SCRIPT_DIR = os.path.dirname(os.path.abspath(__file__))
 SYSTEM_KWIN_HELPER = "/usr/local/bin/reforger-kwin-capture"
@@ -748,12 +744,101 @@ def build_templates(font, pointsize):
     )
 
 
+def validate_template_grid(grid, digit, sample_index, path):
+    if not isinstance(grid, list) or len(grid) != 36:
+        sys.exit(f"invalid template grid for digit {digit} sample {sample_index} in {path}: expected 36 rows")
+    for row in grid:
+        if not isinstance(row, list) or len(row) != 24 or any(not isinstance(value, bool) for value in row):
+            sys.exit(
+                f"invalid template grid for digit {digit} sample {sample_index} in {path}: "
+                "expected 24 boolean columns per row"
+            )
+
+
+def load_template_set(path):
+    try:
+        with open(path, "r", encoding="utf-8") as handle:
+            payload = json.load(handle)
+    except FileNotFoundError:
+        sys.exit(f"missing digit template set {path}; regenerate it from the regression dataset")
+    except json.JSONDecodeError as exc:
+        sys.exit(f"failed to parse digit template set {path}: {exc}")
+
+    digits = payload.get("digits") if isinstance(payload, dict) else None
+    if not isinstance(digits, dict):
+        sys.exit(f"invalid digit template set {path}: missing digits object")
+
+    templates = {}
+    for digit in "0123456789":
+        samples = digits.get(digit)
+        if not isinstance(samples, list) or not samples:
+            sys.exit(f"invalid digit template set {path}: missing samples for digit {digit}")
+        templates[digit] = []
+        for index, grid in enumerate(samples):
+            validate_template_grid(grid, digit, index, path)
+            templates[digit].append(grid)
+
+    return templates
+
+
+def templates_from_dataset(paths, reference_crop, reference_size, scale_mode, cropped):
+    templates = {str(digit): [] for digit in range(10)}
+    for path in paths:
+        expected = os.path.splitext(os.path.basename(path))[0]
+        if not expected.isdigit():
+            sys.exit(f"template source filename must be numeric: {path}")
+
+        crop = resolve_crop(path, None, reference_crop, reference_size, scale_mode, cropped)
+        width, height, raw = load_queue_image(path, crop, cropped)
+        components = connected_components(orange_digit_mask(width, height, raw))
+        if len(components) != len(expected):
+            sys.exit(
+                f"template source {path} produced {len(components)} digit components, "
+                f"but filename expects {len(expected)}"
+            )
+
+        for digit, component in zip(expected, components):
+            templates[digit].append(normalize(component))
+
+    missing = [digit for digit, samples in templates.items() if not samples]
+    if missing:
+        sys.exit(f"template dataset has no samples for digits: {', '.join(missing)}")
+    return templates
+
+
+def write_template_set(output, templates, source_paths):
+    payload = {
+        "version": 1,
+        "normalize_size": [24, 36],
+        "source": "regression digit crops",
+        "source_files": [os.path.basename(path) for path in source_paths],
+        "digits": templates,
+    }
+    directory = os.path.dirname(output)
+    if directory:
+        os.makedirs(directory, exist_ok=True)
+    temp_path = f"{output}.tmp"
+    with open(temp_path, "w", encoding="utf-8") as handle:
+        json.dump(payload, handle, separators=(",", ":"))
+        handle.write("\n")
+    os.replace(temp_path, output)
+
+
+def resolve_templates(template_set, font, pointsize):
+    if font:
+        return build_templates(font, pointsize)
+    return load_template_set(template_set)
+
+
 def classify(component, templates):
     sample = normalize(component)
     ranked = []
-    for digit, template in templates.items():
-        hamming, iou = template_distance(sample, template)
-        ranked.append((hamming, -iou, digit, iou))
+    for digit, digit_templates in templates.items():
+        if digit_templates and isinstance(digit_templates[0][0], bool):
+            digit_templates = [digit_templates]
+        for template in digit_templates:
+            hamming, iou = template_distance(sample, template)
+            ranked.append((hamming, -iou, digit, iou))
     ranked.sort()
     hamming, negative_iou, digit, iou = ranked[0]
     return digit, hamming, iou
@@ -1213,10 +1298,10 @@ def load_queue_image(image_path, crop, already_cropped=False):
     return ppm_from_magick([image_path, "+repage", "-crop", f"{width}x{height}+{x}+{y}", "+repage"])
 
 
-def read_queue_number(image_path, crop, font, pointsize, already_cropped=False):
+def read_queue_number(image_path, crop, template_set, font, pointsize, already_cropped=False):
     width, height, raw = load_queue_image(image_path, crop, already_cropped)
     components = connected_components(orange_digit_mask(width, height, raw))
-    templates = build_templates(font, pointsize)
+    templates = resolve_templates(template_set, font, pointsize)
 
     digits = []
     details = []
@@ -1282,6 +1367,16 @@ def main():
     parser = argparse.ArgumentParser(description="Read Arma Reforger queue number from a fixed screen crop.")
     parser.add_argument("--image", help="Read from an existing image.")
     parser.add_argument("--dataset", help="Read every image in a dataset directory.")
+    parser.add_argument(
+        "--build-template-set",
+        metavar="DATASET",
+        help=argparse.SUPPRESS,
+    )
+    parser.add_argument(
+        "--template-output",
+        default=DEFAULT_TEMPLATE_SET,
+        help=argparse.SUPPRESS,
+    )
     parser.add_argument("--list-screens", action="store_true", help="List KWin output names for use with --screen.")
     parser.add_argument(
         "--list-kwin-windows",
@@ -1373,7 +1468,12 @@ def main():
         default="width",
         help="How to scale --reference-crop. Width mode preserves a 16:9 UI layout and is the default.",
     )
-    parser.add_argument("--font", default=DEFAULT_FONT, help="Font used for synthetic digit templates.")
+    parser.add_argument(
+        "--template-set",
+        default=DEFAULT_TEMPLATE_SET,
+        help="Digit template JSON used by default recognition.",
+    )
+    parser.add_argument("--font", help="Render synthetic digit templates from an ImageMagick font instead of using --template-set.")
     parser.add_argument("--pointsize", type=int, default=DEFAULT_POINTSIZE, help="Point size used for templates.")
     parser.add_argument("--debug", action="store_true", help="Print component and match details.")
     parser.add_argument("--show-crop", action="store_true", help="Print the resolved crop used for each full-size image.")
@@ -1406,6 +1506,7 @@ def main():
         for value in (
             args.image,
             args.dataset,
+            args.build_template_set,
             args.capture,
             args.list_screens,
             args.list_kwin_windows,
@@ -1427,6 +1528,21 @@ def main():
         print_kwin_windows(kwin_windows())
         return 0
 
+    if args.build_template_set:
+        extensions = {".png", ".jpg", ".jpeg", ".webp", ".bmp"}
+        paths = [
+            os.path.join(args.build_template_set, name)
+            for name in sorted(os.listdir(args.build_template_set))
+            if os.path.splitext(name)[1].lower() in extensions
+        ]
+        if not paths:
+            print("no template source images found")
+            return 2
+        templates = templates_from_dataset(paths, args.reference_crop, args.reference_size, args.scale_mode, args.cropped)
+        write_template_set(args.template_output, templates, paths)
+        print(args.template_output)
+        return 0
+
     if args.dataset:
         extensions = {".png", ".jpg", ".jpeg", ".webp", ".bmp"}
         paths = [
@@ -1445,7 +1561,7 @@ def main():
             crop = resolve_crop(path, args.crop, args.reference_crop, args.reference_size, args.scale_mode, args.cropped)
             if args.show_crop and crop:
                 print(f"{path}\tcrop={crop[0]},{crop[1]},{crop[2]},{crop[3]}")
-            number, details = read_queue_number(path, crop, args.font, args.pointsize, args.cropped)
+            number, details = read_queue_number(path, crop, args.template_set, args.font, args.pointsize, args.cropped)
             if not number:
                 status = 2
                 failed += 1
@@ -1529,7 +1645,7 @@ def main():
         crop = resolve_crop(image_path, args.crop, args.reference_crop, args.reference_size, args.scale_mode, args.cropped)
         if args.show_crop and crop:
             print(f"crop={crop[0]},{crop[1]},{crop[2]},{crop[3]}")
-        number, details = read_queue_number(image_path, crop, args.font, args.pointsize, args.cropped)
+        number, details = read_queue_number(image_path, crop, args.template_set, args.font, args.pointsize, args.cropped)
     finally:
         if temp_path:
             try: