Original self.io_module is a dict composed of {'node. target': node}. When quantize model has shared convolution layer such as shared detector head, the 'node. target' will matches many nodes. However {'node. target': node} can not record all these situations. Here self.io_module is modified as {'node. target': [node1, node2, .....]} to fix the problem.
Original self.io_module is a dict composed of {'node. target': node}. When quantize model has shared convolution layer such as shared detector head, the 'node. target' will matches many nodes. However {'node. target': node} can not record all these situations. Here self.io_module is modified as {'node. target': [node1, node2, .....]} to fix the problem.