convert model cost time and consuming memory too much

[chennuo0125-HIT]

test platform : jetson agx xavier memory capacity : 32G os : jetpack4.3 tensorflow : 1.15

i optimize the https://github.com/YunYang1994/tensorflow-yolov3 project with tf_trt, which cost about 15 minutes to load and convert model to optimized model, besides the memory consumed about 26G, it's terriable !!! then i saved optimized model, and then load the optimized model directly without convert, but the result is same. please give me some advise, thanks

[chennuo0125-HIT]

the configure param is :

    converter = trt.TrtGraphConverter(input_graph_def=frozen_graph_def,\
            nodes_blacklist=return_elements,\
            is_dynamic_op=True,\
            precision_mode=trt.TrtPrecisionMode.FP16,\
            )