object_detection export_inference_graph.py --input_type=image_tensor --pipeline_config_path=training/pipeline.config --trained_checkpoint_prefix=training/saved_model/model.ckpt-53958 --output_directory=inference_graph