Rust computer vision and deep learning framework
Custom work-stealing thread pool with priority-aware cold-idle spin — drop-in for rayon::ThreadPool in yscv inference paths
CPU and GPU compute backends with SIMD dispatch and BLAS integration
TOML-driven pipeline configuration + multi-accelerator dispatch for FPV/edge deployments
Post-training quantization CLI: calibrate fp32 ONNX + sample inputs into a QDQ-format INT8 model
CLI for inference, benchmarking, and evaluation
ONNX runtime with 128+ operators, quantization, and graph optimization
Rockchip MPP (Media Process Platform) hardware H.264/HEVC encoder via dlopen
Object detection pipeline with YOLOv8, NMS, and heatmap decoding
Dynamic computation graph with tape-based reverse-mode autodiff
Evaluation metrics (mAP, MOTA, HOTA) and dataset adapters
100+ SIMD-accelerated image processing operations for u8 and f32