affordance-2021

This repository contains a demo of the proposed affordance network. The pretrained network is provided for easy usage. The network was trained using the foreground of the object. Accordingly, input images are expected to contain the object in the foreground.

Table of contents

• Python project
• References

Python project

Requirements

The requirements to run the python code are the following:

• Python 3.7 (64-bit)
• Tensorflow 2.X
• OpenCV

Description

There are 3 folders:

• models: holds the main model for affordance detection (MobileNetV1_UNET) in TFLite format.
• images: holds some images used during the inference phase.
• script: holds two python files. data_loader.py consists of methods to perform image processing operations. affordance_inference_tflite.py performs the affordance prediction.

References

Some (hopefully) useful links:

• TFLite
• TensorFlow Lite for Android