Fig. 1. Module of ReActNet

Fig. 2. (a) Input image；(b) Input features of convolution layer；(c) Binarized features； (d) Reconstructed features after convolution

Fig. 3. (a) Block diagram of network; (b) Block diagram of optimized convolution module

Fig. 4. (a) Input image； (b) Feature intensity at the yellow line

Fig. 5. (a) Input image；(b) Input features of convolution layer；(c) Binarized features of ARSign pathway；(d) Binarized features of RSign pathway；(e) Reconstructed features after convolution

Fig. 6. ReActFace test results in FDDB test set (Baseline refers to the benchmark network, Float32 Model refers to the full precision SSD network)

Fig. 7. Module ablation results in FDDB validation set test

Fig. 8. Result on NIR Images