ZQCNN is a lightweight CNN inference runtime written in C/C++ that supports cross-platform deployment including Windows, Linux, and ARM-Linux. It provides optimized convolution implementations using SIMD instructions (SSE/AVX for x86, NEON for ARM) and integrates with BLAS libraries like Intel MKL and OpenBLAS. The framework includes demo applications for face detection and recognition tasks, and is designed to work with models converted from TensorFlow and trained via PyTorch pipelines.