Proposed deep neural network architecture. Panel a: shows our proposed channel sharing spatio-temporal network (CSSTN). Panel b: is a channel sharing spatio-temporal filter adaptive network (CSSTFAN) sub-network. Panel c: is the meaning of different color blocks and symbols in the network. CSSTN consists of three sub-networks: a feature extraction network, a CSSTFAN, and a reconstruction network. Firstly the feature extraction network extracts features E_t from the current blurry image B_t. Given the blurred image B_t − 1 and restored image R_t − 1 of the previous time step, and current input image B_t, the CSSTFAN generates the F_svi and F_svc in order. Using CSFAC layer ⊛, CSSTFAN convolves F_svi with features H_t − 1 of the previous time step and convolves F_svc with features E_t. Finally, the reconstruction network restores the latent image from the fused features C_t. k denotes the filter size of CSFAC layer and it is 3 in our final network.