Pro Tips & Optimization

Accuracy is the foundation of a successful RaceTagger workflow. Here are the key strategies to get the best possible detections:

• Choose the specific sport category instead of Auto Detect when possible. The AI model is optimized for each sport, so manually selecting your sport significantly improves accuracy.
• Use a participant preset — it helps the AI correct ambiguous detections and improves confidence scores.
• Shoot from optimal angles: Side-angle shots work best for cars; front/back angles are ideal for runners and cyclists.
• Lighting matters. Well-lit photos with visible, high-contrast numbers consistently produce the best results.
• Avoid problem scenarios: The AI struggles with heavily obscured numbers, extreme motion blur, and very low-resolution photos. Consider re-shooting if critical photos fall into these categories.

Processing large photo libraries can take time, but several factors can speed things up significantly:

• JPEG files process faster than RAW. JPEGs skip the extraction step and go straight to AI analysis, while RAW files require embedded preview extraction first.
• Adjust the optimization level in Settings to match your hardware:
  • Conservative: 8 workers, safe for older machines or limited RAM
  • Balanced: 12 workers (default), good performance for most systems
  • Aggressive: 20 workers, fast processing but requires 16GB+ RAM
• Close other applications while processing large batches. Heavy applications (video editors, IDEs, web browsers with many tabs) consume CPU and memory, slowing down RaceTagger.
• Use SSD storage. SSDs significantly speed up RAW processing compared to mechanical drives, especially when reading large files and extracting metadata.

Understanding token consumption helps you budget your analysis work and get the most value from your tokens:

• 1 token = 1 image analyzed, regardless of file format, size, or dimensions. A 5MB RAW file and a 2MB JPEG both cost exactly 1 token.
• Every new account gets 1,500 free tokens to get started. That's plenty for testing and initial events.
• Tokens never expire. You can use them whenever you want, so there's no pressure to rush through your library.
• Monitor your balance in the RaceTagger app's Settings menu or at racetagger.cloud/account.
• Process a test batch first. Analyze 10-20 photos to verify your settings, sport category, and preset before running through thousands of photos.
• Failed analyses don't consume tokens. If a photo fails due to network error or corruption, you won't be charged.

RaceTagger is fully RAW-capable. Here's how to work with your RAW files efficiently:

Supported formats: NEF (Nikon), ARW (Sony), CR2/CR3 (Canon), ORF (Olympus), DNG, RW2 (Panasonic).

• RAW files are processed by extracting the embedded JPEG preview. This approach is fast and preserves quality — we're not re-encoding, just using the preview the camera already created.
• Metadata is written to XMP sidecar files by default, keeping your RAW files completely untouched. This non-destructive approach lets you maintain archival integrity.
• Import both RAW and XMP files into Lightroom to see the detected numbers and metadata alongside your images. Lightroom will automatically read the XMP sidecars.

Processing large event photo collections requires a bit of planning for best results:

• Organize photos into a single folder before starting. This makes batch processing straightforward and keeps results organized.
• Remove obviously irrelevant photos first (test shots, setup photos, blurry test images, duplicate angles). This saves tokens and keeps results clean.
• For multi-day events, process each day as a separate batch. This lets you adjust settings per day (different lighting, venues, or sport categories) and makes organizing results easier.
• You can pause and resume processing at any time. Don't feel pressured to run the entire batch in one sitting.
• The app saves progress automatically. If RaceTagger crashes or you force-quit, launch it again and you can resume from where you left off — no photos are re-analyzed.

Extracting value from your analysis results goes beyond just reviewing the output:

• Review low-confidence detections manually. These often contain correctly detected numbers that just scored lower on confidence. A quick visual check can recover valuable matches.
• Use the "Unknown" folder as a quality check. Browse through it to catch any numbers the AI may have missed. This helps you identify lighting or angle issues for future events.
• Export results as CSV for record-keeping, sharing with event organizers, or further processing in a spreadsheet.
• Photos with multiple detected numbers are automatically copied to all relevant result folders, making it easy to review multi-athlete shots.

Speed up your workflow with these handy shortcuts and quick actions: