usage_instructions.md

Usage Instructions

The scripts should be run in the following order. See the models directory for some examples.

To create the optimized inference graph,

1. If the input is a saved TensorFlow checkpoint or SavedModel, run create_inference_graph.py. You can skip this if you have the inference graph already.
2. To optimize an inference graph, run optimize_inference_graph.py

To generate a design from the TensorFlow model,

1. Run dnn_to_spatial.py and pass the inference graph as input
2. Follow the printed instructions to run the Spatial compiler for the generated application
3. Once Spatial compilation finishes, follow the aws instructions to load the generated AFI to your EC2 F1 instance and run the inference

To give input data to Spatial,

1. Run data/img_to_csv.py on the input image
2. Pass the .csv file as an argument to the Spatial Top executable