tensorflow-resnet50-example

Introduction

This repository is dedicated to provide source code for training :

• Build and run a resnet50 on MPPA(R) from scratch

External package

A partial validation dataset (at least 1000 images) of ImageNet (ILSVRC-2012) is required for the calibration at quantization step. You need to register on http://www.image-net.org/download-images in order to get the link to download the dataset.

Contents

You should find for the training:

• images dir: some images to test inference
• network.yaml: basic configuration file to generate model with KaNN(TM)

Python scripts as tensorflow tools:

• import_frozen_model_to_tensorboard.py : python script to create a log DIR for tensorboard visualization
• summary_graph.py: parse nodes and count the number of nodes in pb graph

Preliminary scripts to build models:

• resnet50.py: build a resnet50(imagenet) saved model in tensorflow 2.1
• freeze_model.py: freeze a saved model in protobuf file (.pb)
• convert_tf_to_tflite.py: convert a saved mode to tflite with calibration (required: ILSVRC2012 validation set)

Scripts to run inference with models:

• run_saved_model_inference.py: load and get prediction from a saved_model
• run_frozen_model_inference.py: load and get prediction from a frozen graph (pb)
• run_tflite_inference.py: load and get prediction from a tflite model

Requirements

• CPU (x86) / K200 PCIe (ACE-acceleration)
• MPPA(R) Accesscore release ACE_4.6.0
  • KaNN(TM), 4.6.0