Instructions to use Emreuludasdemir/teknofest2026-task3 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- LightGlue
How to use Emreuludasdemir/teknofest2026-task3 with LightGlue:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
File size: 15,598 Bytes
699f3cd | 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 | from __future__ import annotations
"""Standalone Task 3 reference modality detection.
Decision policy:
1. EXIF wins when it provides an explicit capture-mode signal.
2. DJI `ImageDescription=default` is treated as an explicit RGB-mode marker for
the official 2026 M3T reference set.
3. Thermal keywords such as `WhiteHot` are treated as explicit thermal markers.
4. When EXIF is absent or inconclusive, pixel analysis is used:
- low saturation + high inter-channel correlation suggest grayscale content
- a bimodal histogram strengthens the thermal hypothesis
- higher saturation + less-correlated channels suggest RGB content
Named thresholds are kept at module scope so follow-up calibration can adjust
them without rewriting detection logic.
"""
from enum import Enum
from pathlib import Path
from typing import Any
import numpy as np
from PIL import ExifTags, Image, UnidentifiedImageError
from src.core.vision import cv2, is_cv2_available
MAX_ANALYSIS_LONG_SIDE = 640
HSV_GRAY_SATURATION_MAX = 10.0
HSV_RGB_SATURATION_MIN = 20.0
CHANNEL_CORRELATION_GRAY_MIN = 0.98
CHANNEL_CORRELATION_RGB_MAX = 0.97
HISTOGRAM_BIN_COUNT = 32
HISTOGRAM_PEAK_MIN_FRACTION = 0.015
HISTOGRAM_MIN_PEAK_SEPARATION_BINS = 4
HISTOGRAM_DIP_RATIO_MAX = 0.60
HISTOGRAM_SPREAD_MIN_FRACTION = 0.18
DYNAMIC_RANGE_MIN = 40.0
THERMAL_TEXT_KEYWORDS = ("whitehot", "blackhot", "hothot", "thermal", "infrared", "lwir", "mwir")
THERMAL_EXACT_SHORT_KEYWORDS = {"ir"}
RGB_TEXT_KEYWORDS = ("default",)
THERMAL_ONLY_MODEL_KEYWORDS: tuple[str, ...] = ()
EXIF_TEXT_FIELDS = {
"ImageDescription",
"UserComment",
"XPComment",
"XPSubject",
"XPTitle",
"Software",
"Make",
"Model",
}
WEAK_RGB_EXIF_FIELDS = {"WhiteBalance", "Saturation", "ColorSpace"}
class Modality(Enum):
RGB = "rgb"
THERMAL = "thermal"
UNKNOWN = "unknown"
def detect_modality(image_path: Path) -> tuple[Modality, dict[str, Any]]:
"""Classify a reference image as RGB, thermal, or unknown."""
path = Path(image_path)
if not path.exists():
raise FileNotFoundError(path)
if not path.is_file():
raise ValueError(f"Image path is not a file: {path}")
if path.stat().st_size <= 0:
raise ValueError(f"Image file is empty: {path}")
try:
with Image.open(path) as image:
image.load()
exif_info = _extract_exif_info(image)
rgb_image = image.convert("RGB")
except UnidentifiedImageError as exc:
raise ValueError(f"Invalid image file: {path}") from exc
except OSError as exc:
raise ValueError(f"Unable to read image file: {path}") from exc
pixel_signals = _analyze_pixels(np.asarray(rgb_image, dtype=np.uint8))
diagnostics: dict[str, Any] = {
"method": "none",
"exif_signals": list(exif_info["signals"]),
"pixel_signals": pixel_signals,
"confidence": "low",
"reason": "No modality signal found.",
}
if exif_info["strong_thermal"]:
diagnostics.update(
{
"method": "exif",
"confidence": "high",
"reason": f"Thermal EXIF signal matched: {', '.join(exif_info['signals'])}.",
}
)
return Modality.THERMAL, diagnostics
if exif_info["strong_rgb"]:
diagnostics.update(
{
"method": "exif",
"confidence": "high",
"reason": f"RGB EXIF signal matched: {', '.join(exif_info['signals'])}.",
}
)
return Modality.RGB, diagnostics
pixel_modality, pixel_confidence, pixel_reason = _classify_from_pixel_signals(pixel_signals)
if exif_info["weak_rgb"] and pixel_modality == Modality.RGB:
diagnostics.update(
{
"method": "both",
"confidence": "high",
"reason": f"Weak RGB EXIF evidence agrees with pixel analysis: {pixel_reason}",
}
)
return Modality.RGB, diagnostics
if exif_info["weak_rgb"]:
diagnostics.update(
{
"method": "exif",
"confidence": "medium",
"reason": "Weak RGB EXIF evidence was present but pixel analysis was inconclusive or conflicting.",
}
)
return Modality.RGB, diagnostics
diagnostics.update(
{
"method": "pixel",
"confidence": pixel_confidence,
"reason": pixel_reason,
}
)
return pixel_modality, diagnostics
def _extract_exif_info(image: Image.Image) -> dict[str, Any]:
signals: list[str] = []
strong_thermal = False
strong_rgb = False
weak_rgb = False
try:
exif = image.getexif()
except Exception:
exif = None
if not exif:
return {
"signals": signals,
"strong_thermal": strong_thermal,
"strong_rgb": strong_rgb,
"weak_rgb": weak_rgb,
}
tag_names = ExifTags.TAGS
text_fields: dict[str, str] = {}
for raw_key, raw_value in exif.items():
tag_name = str(tag_names.get(raw_key, raw_key))
value_text = _coerce_exif_value_to_text(raw_value)
if tag_name in EXIF_TEXT_FIELDS and value_text:
text_fields[tag_name] = value_text
if tag_name in WEAK_RGB_EXIF_FIELDS:
numeric_value = _coerce_exif_value_to_number(raw_value)
if tag_name == "Saturation" and numeric_value is not None and numeric_value != 0:
weak_rgb = True
signals.append("saturation")
elif tag_name in {"WhiteBalance", "ColorSpace"}:
weak_rgb = True
signals.append(tag_name.lower())
for field_name, value_text in text_fields.items():
lowered = value_text.lower()
if any(keyword in lowered for keyword in RGB_TEXT_KEYWORDS):
strong_rgb = True
signals.append("default")
for keyword in THERMAL_TEXT_KEYWORDS:
if keyword in lowered:
strong_thermal = True
signals.append(keyword)
tokenized = {token.strip(" ,;:_-") for token in lowered.split()}
if any(keyword in tokenized for keyword in THERMAL_EXACT_SHORT_KEYWORDS):
strong_thermal = True
signals.extend(sorted(THERMAL_EXACT_SHORT_KEYWORDS.intersection(tokenized)))
if field_name == "Model" and any(keyword in lowered for keyword in THERMAL_ONLY_MODEL_KEYWORDS):
strong_thermal = True
signals.append("thermal_model")
deduped_signals: list[str] = []
for signal in signals:
if signal not in deduped_signals:
deduped_signals.append(signal)
return {
"signals": deduped_signals,
"strong_thermal": strong_thermal,
"strong_rgb": strong_rgb,
"weak_rgb": weak_rgb,
}
def _analyze_pixels(rgb_image: np.ndarray) -> dict[str, Any]:
analysis_image = _resize_for_analysis(rgb_image)
mean_saturation = _compute_mean_saturation(analysis_image)
correlations = _compute_channel_correlations(analysis_image)
min_channel_correlation = min(correlations.values()) if correlations else 1.0
gray = _to_grayscale(analysis_image)
histogram_info = _compute_histogram_signals(gray)
p5, p95 = np.percentile(gray, [5, 95])
dynamic_range = float(p95 - p5)
return {
"analysis_shape": [int(analysis_image.shape[0]), int(analysis_image.shape[1])],
"mean_saturation": round(mean_saturation, 4),
"channel_correlations": {key: round(value, 6) for key, value in correlations.items()},
"min_channel_correlation": round(float(min_channel_correlation), 6),
"histogram_bimodal": bool(histogram_info["bimodal"]),
"histogram_peak_bins": list(histogram_info["peak_bins"]),
"histogram_peak_values": [round(float(value), 6) for value in histogram_info["peak_values"]],
"histogram_dip_ratio": round(float(histogram_info["dip_ratio"]), 6),
"histogram_spread_fraction": round(float(histogram_info["spread_fraction"]), 6),
"dynamic_range_p5_p95": round(dynamic_range, 4),
}
def _classify_from_pixel_signals(pixel_signals: dict[str, Any]) -> tuple[Modality, str, str]:
mean_saturation = float(pixel_signals["mean_saturation"])
min_corr = float(pixel_signals["min_channel_correlation"])
bimodal = bool(pixel_signals["histogram_bimodal"])
spread_fraction = float(pixel_signals["histogram_spread_fraction"])
dynamic_range = float(pixel_signals["dynamic_range_p5_p95"])
near_grayscale = mean_saturation < HSV_GRAY_SATURATION_MAX and min_corr >= CHANNEL_CORRELATION_GRAY_MIN
clearly_color = mean_saturation >= HSV_RGB_SATURATION_MIN and min_corr <= CHANNEL_CORRELATION_GRAY_MIN
if near_grayscale and bimodal:
return (
Modality.THERMAL,
"high",
(
"Pixel analysis indicates thermal content: low saturation, very high channel correlation, "
"and a bimodal grayscale histogram."
),
)
if near_grayscale and dynamic_range >= DYNAMIC_RANGE_MIN:
return (
Modality.THERMAL,
"medium",
"Pixel analysis indicates grayscale / thermal-like content with broad dynamic range.",
)
if clearly_color and spread_fraction >= HISTOGRAM_SPREAD_MIN_FRACTION:
return (
Modality.RGB,
"high",
"Pixel analysis indicates RGB content with meaningful saturation and non-grayscale channels.",
)
if clearly_color:
return (
Modality.RGB,
"medium",
"Pixel analysis indicates RGB-like channel diversity, but histogram spread is limited.",
)
return (
Modality.UNKNOWN,
"low",
"Pixel analysis produced mixed signals; modality left unknown.",
)
def _resize_for_analysis(rgb_image: np.ndarray) -> np.ndarray:
height, width = rgb_image.shape[:2]
long_side = max(height, width)
if long_side <= MAX_ANALYSIS_LONG_SIDE:
return rgb_image
scale = MAX_ANALYSIS_LONG_SIDE / float(long_side)
new_width = max(1, int(round(width * scale)))
new_height = max(1, int(round(height * scale)))
if is_cv2_available():
return cv2.resize(rgb_image, (new_width, new_height), interpolation=cv2.INTER_AREA)
return np.asarray(Image.fromarray(rgb_image).resize((new_width, new_height), Image.Resampling.BILINEAR))
def _compute_mean_saturation(rgb_image: np.ndarray) -> float:
if is_cv2_available():
hsv = cv2.cvtColor(rgb_image, cv2.COLOR_RGB2HSV)
return float(hsv[..., 1].mean())
hsv = np.asarray(Image.fromarray(rgb_image).convert("HSV"), dtype=np.uint8)
return float(hsv[..., 1].mean())
def _compute_channel_correlations(rgb_image: np.ndarray) -> dict[str, float]:
correlations: dict[str, float] = {}
channels = [rgb_image[..., index].astype(np.float32).reshape(-1) for index in range(3)]
labels = [("r_g", 0, 1), ("r_b", 0, 2), ("g_b", 1, 2)]
for label, left_idx, right_idx in labels:
left = channels[left_idx]
right = channels[right_idx]
left_std = float(np.std(left))
right_std = float(np.std(right))
left_mean = float(np.mean(left))
right_mean = float(np.mean(right))
if left_std < 1e-6 and right_std < 1e-6:
correlations[label] = 1.0 if abs(left_mean - right_mean) < 1.0 else 0.0
continue
if left_std < 1e-6 or right_std < 1e-6:
correlations[label] = 0.0
continue
correlations[label] = float(np.corrcoef(left, right)[0, 1])
return correlations
def _to_grayscale(rgb_image: np.ndarray) -> np.ndarray:
if is_cv2_available():
return cv2.cvtColor(rgb_image, cv2.COLOR_RGB2GRAY)
red = rgb_image[..., 0].astype(np.float32)
green = rgb_image[..., 1].astype(np.float32)
blue = rgb_image[..., 2].astype(np.float32)
return np.clip((0.299 * red) + (0.587 * green) + (0.114 * blue), 0, 255).astype(np.uint8)
def _compute_histogram_signals(gray_image: np.ndarray) -> dict[str, Any]:
histogram, _ = np.histogram(gray_image, bins=HISTOGRAM_BIN_COUNT, range=(0, 256))
histogram = histogram.astype(np.float64)
histogram_sum = float(histogram.sum())
if histogram_sum <= 0.0:
return {
"bimodal": False,
"peak_bins": [],
"peak_values": [],
"dip_ratio": 1.0,
"spread_fraction": 0.0,
}
histogram /= histogram_sum
spread_fraction = float(np.count_nonzero(histogram > HISTOGRAM_PEAK_MIN_FRACTION) / float(HISTOGRAM_BIN_COUNT))
peaks: list[tuple[int, float]] = []
for index, value in enumerate(histogram):
left = histogram[index - 1] if index > 0 else -1.0
right = histogram[index + 1] if index + 1 < len(histogram) else -1.0
if value >= left and value >= right and value >= HISTOGRAM_PEAK_MIN_FRACTION:
peaks.append((index, float(value)))
best_pair: tuple[tuple[int, float], tuple[int, float]] | None = None
best_pair_score = -1.0
for left_index in range(len(peaks)):
for right_index in range(left_index + 1, len(peaks)):
left_peak = peaks[left_index]
right_peak = peaks[right_index]
separation = abs(left_peak[0] - right_peak[0])
if separation < HISTOGRAM_MIN_PEAK_SEPARATION_BINS:
continue
pair_score = min(left_peak[1], right_peak[1])
if pair_score > best_pair_score:
best_pair = (left_peak, right_peak)
best_pair_score = pair_score
if best_pair is None:
top_peaks = sorted(peaks, key=lambda item: item[1], reverse=True)[:2]
return {
"bimodal": False,
"peak_bins": [item[0] for item in top_peaks],
"peak_values": [item[1] for item in top_peaks],
"dip_ratio": 1.0,
"spread_fraction": spread_fraction,
}
left_peak, right_peak = sorted(best_pair, key=lambda item: item[0])
valley = histogram[left_peak[0] : right_peak[0] + 1]
valley_floor = float(valley.min()) if len(valley) else 1.0
smaller_peak = min(left_peak[1], right_peak[1])
dip_ratio = valley_floor / max(smaller_peak, 1e-9)
return {
"bimodal": dip_ratio <= HISTOGRAM_DIP_RATIO_MAX,
"peak_bins": [left_peak[0], right_peak[0]],
"peak_values": [left_peak[1], right_peak[1]],
"dip_ratio": float(dip_ratio),
"spread_fraction": spread_fraction,
}
def _coerce_exif_value_to_text(value: Any) -> str:
if isinstance(value, bytes):
for encoding in ("utf-8", "utf-16", "latin1"):
try:
return value.decode(encoding, errors="ignore").strip("\x00 ").strip()
except Exception:
continue
return ""
if isinstance(value, (tuple, list)):
return " ".join(_coerce_exif_value_to_text(item) for item in value if _coerce_exif_value_to_text(item))
if value is None:
return ""
return str(value).strip()
def _coerce_exif_value_to_number(value: Any) -> float | None:
if isinstance(value, (int, float)):
return float(value)
if isinstance(value, (tuple, list)) and len(value) == 2 and all(isinstance(item, (int, float)) for item in value):
denominator = float(value[1])
if denominator == 0.0:
return None
return float(value[0]) / denominator
return None
|