edge_vision_core 0.1.0
edge_vision_core: ^0.1.0 copied to clipboard
Deterministic C++17 image & camera frame preprocessor, EXIF orienter, letterbox resizer, tensor normalizer, and YOLO postprocessor for Flutter Edge AI.
0.1.0 #
- Initial release of
edge_vision_core. - Shared C++17 core (
src/edge_vision_core.cpp,src/exif.cpp) with a cleanextern "C"ABI (src/edge_vision_core.h) reusable across Flutter, React Native, and Python. - Built-in JPEG
APP1TIFF/IFD0 EXIF orientation parser (1..8, little-endian and big-endian) and physical pixel rotation/flip normalization. - Deterministic
letterbox(114 padding),stretch, andcenterCropresizing across Android, iOS, Windows, macOS, and Linux. - Single-pass
RGB/BGR,NCHW/NHWC,scale/mean/stdDevfloat32 tensor normalization with presets forYOLO,ImageNet, andCLIP. - Live camera
YUV420/NV12/NV21deterministic BT.601 conversion with clockwise sensor rotation (0,90,180,270). - EXIF-free resized JPEG encoding (
resizeAndEncodeJpeg) for deterministic cloud VLM / LLM Vision upload reduction. - Native YOLO post-processor (
postprocessYolo) supporting both classic[1, 4 + C, N](with class-aware Greedy NMS) and end-to-end NMS-free[1, N, 6](YOLO26 / YOLOv10) outputs, automatically unscaling boxes back to oriented source coordinates. - Thread-safe serialized Dart FFI queue and background
Isolate.runhelper (preprocessInIsolate).