Numera

An app for batch processing and numbering photos.
Based on some features of FastStone Image Viewer.
Created using the following xtask-template.

Features

• Able to convert images in batch to pdf/jpg
• Able to add text to the images (e.g. automatically add file name in the corner, useful for printing and selling posters)
• Able to rotate the images
• Poster options for 33x66cm centered crop/resize and optional white border margin
• Optional PNG/JPEG image watermark with position, rotation, opacity, and size controls
• Numbering mode for fast per-image number tagging
• OCR-based number suggestions
• Sorting mode to quickly sort photos into subfolders based on their name

Batch processing preview

Numbering mode preview

Sorintg mode preview

OCR setup (optional, Numbering mode)

The app can suggest motorcycle numbers using local ONNX OCR.

1. Put OCR assets in models/ppocrv5/ (or set BIP_OCR_MODEL_DIR).
2. Ensure this folder contains:
   • det.onnx
   • rec.onnx
   • PP-LCNet_x0_25_textline_ori.onnx
   • PP-LCNet_x1_0_doc_ori.onnx
   • ppocrv5_dict.txt

If these files are missing, the app still runs, but OCR suggestions stay disabled.

OCR acceleration device (optional)

OCR can use GPU/accelerator execution providers when compiled with the matching Cargo feature.

Examples:

• NVIDIA CUDA:
  • Build: cargo run --release --features ocr-cuda
  • Run: BIP_OCR_DEVICE=cuda:0 ./target/release/numera
• TensorRT:
  • Build: cargo run --release --features ocr-tensorrt
  • Run: BIP_OCR_DEVICE=tensorrt:0 ./target/release/numera
• DirectML (Windows): feature ocr-directml, BIP_OCR_DEVICE=directml:0
• CoreML (macOS): feature ocr-coreml, BIP_OCR_DEVICE=coreml
• OpenVINO: feature ocr-openvino, BIP_OCR_DEVICE=openvino:GPU
• WebGPU: feature ocr-webgpu, BIP_OCR_DEVICE=webgpu

Defaults:

• BIP_OCR_DEVICE=cpu (default)
• BIP_OCR_THREADS controls OCR intra-op threads. If unset, the app auto-detects CPU parallelism but caps the default at 8 threads so OCR does not make preview/navigation sluggish. Set this variable explicitly to override that default, for example BIP_OCR_THREADS=16.

Threading notes:

• intra_threads means threads used inside one ONNX Runtime operation/model node.
• inter_threads means threads used to run independent ONNX Runtime graph operations in parallel.
• The app uses inter_threads=1 because OCR is mostly a small sequential pipeline and higher inter-op parallelism can oversubscribe the CPU without improving latency.
• ONNX Runtime graph optimizations, memory-pattern optimization, parallel execution mode, and CPU fallback are enabled for both CPU and accelerator configurations. GPU/accelerator providers run supported model operators on the device; CPU threads are still used for preprocessing, scheduling, unsupported fallback operators, and image loading.

If an accelerator provider is unavailable at runtime, OCR automatically falls back to CPU.

Preview sizing

Numbering and batch previews are decoded to a bounded maximum side length for responsiveness.

Defaults:

• BIP_PREVIEW_MAX_SIDE=2400 effective default.
• Accepted range is clamped to 512..8192.

Increase it for sharper previews/zooming, for example:

• BIP_PREVIEW_MAX_SIDE=3200 ./target/release/numera

Decrease it if folder navigation is still too slow on your machine, for example:

• BIP_PREVIEW_MAX_SIDE=1600 ./target/release/numera

Sticker template assist (Numbering mode)

You can load an event sticker template from the Numbering toolbar (Sticker Template button). When loaded, OCR first tries to locate that sticker in each photo and prioritizes the number area of that sticker before falling back to full-image OCR. This helps reduce confusion with old-event stickers that appear in the same frame.