I have implemented the LFFD referring to the official python implementation

paper:LFFD: A Light and Fast Face Detector for Edge Devices

official github: LFFD

Also,my ncnn implementation is here

• You can set the input tensor shape smaller ,since you need to reduce the memory and accelerate the inference.
• You can set the scale_num=8 to use another larger model.
• I just test it on vs2019 PC and the result is correct compared to original implementation,you can use the code to another device such as android、RK3399、and so on.

• TensorRT demo: mmxnet model --> onnx-->trt engine
• openvino demo: mxnet model-->onnx-->openvino