Spaces:
Sleeping
Sleeping
File size: 13,599 Bytes
aa2c8ff 1386d71 aa2c8ff 6c65498 aa2c8ff 6c65498 aa2c8ff 6c65498 aa2c8ff | 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 | #!/usr/bin/env python3
"""
Test template matching accuracy against OCR ground truth.
This test:
1. Collects samples from a longer segment to get full digit coverage
2. Splits samples into training (build templates) and test (evaluate accuracy) sets
3. Compares template matching results against OCR ground truth
4. Measures timing improvement
5. Saves debug images for wrong/undetected cases (if <= 10 total errors)
Uses dual-mode matching to handle both single-digit (centered) and double-digit
(left/right) layouts. Templates needed: 25 total (10 center + 10 right + 4 tens + 1 blank).
Usage:
cd /Users/andytaylor/Documents/Personal/cfb40
source .venv/bin/activate
python tests/test_digit_templates/test_template_accuracy.py
"""
import logging
import sys
import time
from pathlib import Path
from typing import List
import cv2
import numpy as np
from detection import DetectScoreBug
from readers import ReadPlayClock
from setup import DigitTemplateBuilder, PlayClockRegionExtractor
logging.basicConfig(level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s")
logger = logging.getLogger(__name__)
# Test configuration
VIDEO_PATH = "full_videos/OSU vs Tenn 12.21.24.mkv"
TEMPLATE_PATH = "output/OSU_vs_Tenn_12_21_24_template.png"
PLAYCLOCK_CONFIG_PATH = "output/OSU_vs_Tenn_12_21_24_playclock_config.json"
# Use longer segment to get more digit coverage
# 38:40 to 48:40 = 10-minute segment with ~13 plays per v3 baseline
START_TIME = 38 * 60 + 40 # 2320 seconds
END_TIME = 48 * 60 + 40 # 2920 seconds
SAMPLE_INTERVAL = 0.5
# Debug output directory
DEBUG_DIR = Path("output/debug/digit_templates/errors")
def collect_all_samples(video_path: str, start_time: float, end_time: float, sample_interval: float):
"""
Collect play clock samples with OCR ground truth.
Returns list of (timestamp, clock_value, region_image, confidence)
"""
scorebug_detector = DetectScoreBug(template_path=TEMPLATE_PATH)
clock_reader = PlayClockRegionExtractor(region_config_path=PLAYCLOCK_CONFIG_PATH)
cap = cv2.VideoCapture(video_path)
if not cap.isOpened():
raise RuntimeError(f"Could not open video: {video_path}")
fps = cap.get(cv2.CAP_PROP_FPS)
start_frame = int(start_time * fps)
end_frame = int(end_time * fps)
frame_skip = int(sample_interval * fps)
logger.info("Collecting samples from %.1fs to %.1fs", start_time, end_time)
# Lock scorebug region
cap.set(cv2.CAP_PROP_POS_FRAMES, start_frame)
ret, frame = cap.read()
if ret:
scorebug_detector.discover_and_lock_region(frame)
logger.info("Scorebug region: %s", scorebug_detector.fixed_region)
samples = []
current_frame = start_frame
cap.set(cv2.CAP_PROP_POS_FRAMES, start_frame)
while current_frame < end_frame:
ret, frame = cap.read()
if not ret:
break
current_time = current_frame / fps
detection = scorebug_detector.detect(frame)
if detection.detected and detection.bbox:
sb_x, sb_y, _, _ = detection.bbox
pc_config = clock_reader.config
pc_x = sb_x + pc_config.x_offset
pc_y = sb_y + pc_config.y_offset
pc_w = pc_config.width
pc_h = pc_config.height
frame_h, frame_w = frame.shape[:2]
if 0 <= pc_x and 0 <= pc_y and pc_x + pc_w <= frame_w and pc_y + pc_h <= frame_h:
region = frame[pc_y : pc_y + pc_h, pc_x : pc_x + pc_w].copy()
reading = clock_reader.read(frame, detection.bbox)
if reading.detected and reading.value is not None:
samples.append((current_time, reading.value, region, reading.confidence))
for _ in range(frame_skip - 1):
cap.grab()
current_frame += frame_skip
cap.release()
return samples
def split_samples(samples: List, train_ratio: float = 0.7):
"""Split samples into training and test sets."""
# Sort by timestamp to ensure temporal split
sorted_samples = sorted(samples, key=lambda x: x[0])
split_idx = int(len(sorted_samples) * train_ratio)
train_samples = sorted_samples[:split_idx]
test_samples = sorted_samples[split_idx:]
return train_samples, test_samples
def save_debug_images(error_results: List[dict], output_dir: Path):
"""
Save debug images for error cases.
Each image shows:
- Original region (scaled up)
- Preprocessed binary image
- Annotation with OCR value, template value, confidence
"""
output_dir.mkdir(parents=True, exist_ok=True)
# Clear previous debug images
for f in output_dir.glob("*.png"):
f.unlink()
for i, result in enumerate(error_results):
timestamp = result["timestamp"]
ocr_value = result["ocr_value"]
template_value = result["template_value"]
confidence = result["confidence"]
status = result["status"]
region = result["region"]
# Scale up the region for visibility (4x)
scale = 4
scaled_region = cv2.resize(region, None, fx=scale, fy=scale, interpolation=cv2.INTER_NEAREST)
# Create a larger canvas with annotation space
canvas_h = scaled_region.shape[0] + 60
canvas_w = max(scaled_region.shape[1], 300)
canvas = np.zeros((canvas_h, canvas_w, 3), dtype=np.uint8)
canvas[:] = (40, 40, 40) # Dark gray background
# Place scaled region at top
x_offset = (canvas_w - scaled_region.shape[1]) // 2
canvas[0 : scaled_region.shape[0], x_offset : x_offset + scaled_region.shape[1]] = scaled_region
# Add annotations
y_text = scaled_region.shape[0] + 20
font = cv2.FONT_HERSHEY_SIMPLEX
font_scale = 0.5
color = (0, 0, 255) if status == "WRONG" else (0, 165, 255) # Red for wrong, orange for undetected
# Status and timestamp
cv2.putText(canvas, f"{status} @ {timestamp:.1f}s", (10, y_text), font, font_scale, color, 1)
# OCR vs Template
y_text += 18
template_str = str(template_value) if template_value is not None else "None"
cv2.putText(canvas, f"OCR: {ocr_value} Template: {template_str} Conf: {confidence:.2f}", (10, y_text), font, font_scale, (200, 200, 200), 1)
# Save with descriptive filename
if status == "WRONG":
filename = f"wrong_{i:02d}_t{timestamp:.0f}s_ocr{ocr_value}_tmpl{template_value}.png"
else:
filename = f"missed_{i:02d}_t{timestamp:.0f}s_ocr{ocr_value}.png"
cv2.imwrite(str(output_dir / filename), canvas)
logger.info("Saved %d debug images to: %s", len(error_results), output_dir)
def test_template_accuracy():
"""Test template matching accuracy against OCR ground truth."""
logger.info("=" * 60)
logger.info("TEST: Template Matching Accuracy vs OCR")
logger.info("=" * 60)
# Check files exist
for path, name in [(VIDEO_PATH, "Video"), (TEMPLATE_PATH, "Template"), (PLAYCLOCK_CONFIG_PATH, "Config")]:
if not Path(path).exists():
logger.error("%s not found: %s", name, path)
return False
# Collect all samples
logger.info("\n[Step 1] Collecting samples with OCR ground truth...")
all_samples = collect_all_samples(VIDEO_PATH, START_TIME, END_TIME, SAMPLE_INTERVAL)
logger.info("Total samples: %d", len(all_samples))
# Split into train/test
logger.info("\n[Step 2] Splitting samples (70% train, 30% test)...")
train_samples, test_samples = split_samples(all_samples, train_ratio=0.7)
logger.info("Training samples: %d", len(train_samples))
logger.info("Test samples: %d", len(test_samples))
# Build templates from training set
logger.info("\n[Step 3] Building templates from training samples...")
builder = DigitTemplateBuilder()
for timestamp, clock_value, region, confidence in train_samples:
builder.add_sample(region, clock_value, timestamp, confidence)
# Coverage with dual-mode templates (center + right positions)
coverage = builder.get_coverage_status()
logger.info("Training coverage (dual-mode):")
logger.info(" Ones (center): %s (missing: %s)", coverage["ones_center"], coverage["ones_center_missing"])
logger.info(" Ones (right): %s (missing: %s)", coverage["ones_right"], coverage["ones_right_missing"])
logger.info(" Tens (left): %s (missing: %s)", coverage["tens"], coverage["tens_missing"])
logger.info(" Blank: %s", "YES" if coverage["has_blank"] else "NO")
library = builder.build_templates(min_samples=2)
lib_status = library.get_coverage_status()
logger.info("Templates built: %d/%d", lib_status["total_have"], lib_status["total_needed"])
# Test template matching on test set
logger.info("\n[Step 4] Testing template matching accuracy...")
template_reader = ReadPlayClock(library)
correct = 0
wrong = 0
undetected = 0
error_results = [] # Store errors with region images for debug
# Also measure timing
template_times = []
for timestamp, ocr_value, region, ocr_confidence in test_samples:
# Template matching
t_start = time.perf_counter()
template_result = template_reader.read(region)
t_template = time.perf_counter() - t_start
template_times.append(t_template)
if template_result.detected and template_result.value is not None:
if template_result.value == ocr_value:
correct += 1
else:
wrong += 1
error_results.append(
{
"timestamp": timestamp,
"ocr_value": ocr_value,
"template_value": template_result.value,
"confidence": template_result.confidence,
"status": "WRONG",
"region": region, # Store region for debug image
}
)
else:
undetected += 1
error_results.append(
{
"timestamp": timestamp,
"ocr_value": ocr_value,
"template_value": None,
"confidence": template_result.confidence,
"status": "UNDETECTED",
"region": region, # Store region for debug image
}
)
total = correct + wrong + undetected
accuracy = correct / total if total > 0 else 0
detection_rate = (correct + wrong) / total if total > 0 else 0
logger.info("\nAccuracy Results:")
logger.info(" Correct: %d (%.1f%%)", correct, 100 * correct / total if total > 0 else 0)
logger.info(" Wrong: %d (%.1f%%)", wrong, 100 * wrong / total if total > 0 else 0)
logger.info(" Undetected: %d (%.1f%%)", undetected, 100 * undetected / total if total > 0 else 0)
logger.info(" Accuracy (correct/total): %.1f%%", accuracy * 100)
logger.info(" Detection rate: %.1f%%", detection_rate * 100)
# Show error details
if error_results:
logger.info("\nError details:")
for r in error_results[:10]:
if r["status"] == "WRONG":
logger.info(" WRONG @ t=%.1fs: OCR=%d, Template=%d, conf=%.2f", r["timestamp"], r["ocr_value"], r["template_value"], r["confidence"])
else:
logger.info(" UNDETECTED @ t=%.1fs: OCR=%d, conf=%.2f", r["timestamp"], r["ocr_value"], r["confidence"])
# Save debug images if <= 10 total errors
if len(error_results) > 0 and len(error_results) <= 10:
logger.info("\n[Step 4.5] Saving debug images for %d errors...", len(error_results))
save_debug_images(error_results, DEBUG_DIR)
elif len(error_results) > 10:
logger.info("\nSkipping debug images: %d errors > 10 threshold", len(error_results))
# Timing comparison
logger.info("\n[Step 5] Timing comparison...")
avg_template_time = sum(template_times) / len(template_times) if template_times else 0
logger.info(" Template matching: %.3fms/frame", avg_template_time * 1000)
logger.info(" EasyOCR (benchmark): ~48.9ms/frame")
logger.info(" Speedup: ~%.0fx", 48.9 / (avg_template_time * 1000) if avg_template_time > 0 else 0)
# Summary
logger.info("\n" + "=" * 60)
logger.info("TEST SUMMARY")
logger.info("=" * 60)
logger.info("Templates built: %d/%d (%.1f%%)", lib_status["total_have"], lib_status["total_needed"], 100 * lib_status["total_have"] / lib_status["total_needed"])
logger.info("Accuracy: %.1f%% (%d/%d correct)", accuracy * 100, correct, total)
logger.info("Detection rate: %.1f%%", detection_rate * 100)
logger.info("Speedup: ~%.0fx faster than OCR", 48.9 / (avg_template_time * 1000) if avg_template_time > 0 else 0)
# Pass criteria: >= 95% accuracy
passed = accuracy >= 0.95 or (accuracy >= 0.90 and lib_status["total_have"] < lib_status["total_needed"])
if passed:
logger.info("\nTEST: PASSED")
else:
logger.info("\nTEST: FAILED (accuracy %.1f%% < 95%%)", accuracy * 100)
# Save library for use in integration tests
output_dir = Path("output/debug/digit_templates")
output_dir.mkdir(parents=True, exist_ok=True)
library.save(str(output_dir))
logger.info("\nTemplates saved to: %s", output_dir)
if len(error_results) > 0 and len(error_results) <= 10:
logger.info("Debug images saved to: %s", DEBUG_DIR)
return passed
if __name__ == "__main__":
success = test_template_accuracy()
sys.exit(0 if success else 1)
|