Fig. 6
From: Global–local multi-stage temporal convolutional network for cataract surgery phase recognition
Overview diagram of the proposed GL-MSTCN. The proposed GL-MSTCN contains a global–local fine-grained features extraction network (TS-Net) and a multi-stage dilated temporal convolutional network (MSTCN). The MSTCN consists of \(S_1\) and \(S_2\), where \(S_1\) and \(S_2\) are composed of parallel dual dilated convolutional layers and single dilated convolutional layers, respectively. \(L_1\) and \(L_2\) denote the loss functions of \(S_1\) and \(S_2\), respectively. FC is a fully connected layer