Poster Abstract

...

Introduction:

The task of converting images to and from tensor formats is crucial for various computer vision tasks, especially those involving Graph Neural Networks (GNNs). However, performing this conversion on distributed Systems increases time efficiency by making the GNN library compatible with distributed architecture. In attempt to do so, the user has to write code for generating tensors for each image type, so the task is to automate this process.

Objective:

The proposed solution is to introduce an Image conversion plugin in the GNN library which has Python embeddings and appropriate APIs for the conversion, enabling users to effortlessly convert images to tensor formats by calling the module in their code and made it easier for researchers and developers to work with image data in GNN-based applications.

Methodology:

The Image dataset which needs to be processed is sprayed onto the system using ‘BLOB Spraying’. A record structure consisting of fields like ‘id’,‘filename’,
and ‘content’ is defined to store the sprayed images. These Images are distributed to multiple nodes in the cluster based on the image Id. The idea is to convert each image data to 1 dimensional array, and extract its data in Byte format using Matplotlib library, then convert this byte buffer to ndarray using the Numpy library. The obtained ndarray can be passed to Np2Tens module defined in the GNN library of HPCC Systems Machine Learning system. The Np2Tens module generates the corresponding tensor of the image; these tensors can be used for ML applications

Presentation

In this Video Recording, NAME Rohhun provides a tour and explanation of his poster content.

Image Conversion Plugin for GNN Module on HPCC Systems

...

Click on the poster for a larger image.

Insert Poster ImageImage Added