TF TRT Integration removes the input nodes for an Object Detection Model

trt_graph = trt.create_inference_graph(
        input_graph_def=g.as_graph_def(),
        outputs=nodenames,
        max_batch_size=1,
        max_workspace_size_bytes=1 << 25,
        precision_mode="FP32",  # TRT Engine precision "FP32","FP16" or "INT8"
        minimum_segment_size=2  # minimum number of nodes in an engine
        )

In the created trt_graph, the input node are removed(for this model, there are multiple input nodes), how can i pass some nodenames, so that the optimization doesn’t remove/fuse them ?

Bug with TF end