[GSoC] 7. Add Video Anomaly Detection Inference to Gradio

[samet-akcay]

Short description

Currently, Anomalib has a Gradio inference interface. However only image inference is supported. This project would add video support to Anomalib's Gradio inference. The following is an overview of the project at a high level:

Enable OpenVINO's asynchronous inference mode in anomalib to accelerate video prediction inference.
Modify Anomalib's Gradio inference interface to support OpenVINO's async inference mode for video prediction.

OPTIONAL
Add a new algorithm for detecting video anomalies to Anomalib. Note that a portion of the workload may overlap with another project, as both require implementing the async inference capabilities of OpenVINO. This would be a great opportunity to learn how to collaborate with other developers and resolve git merge conflicts, which are common in everyday software development.

Expected outcomes

A new Gradio inference UI that accepts both image and video inputs and performs async OpenVINO inference based on a video anomaly detection model that is optionally added by the applicant.

Skills required/preferred

DL (Elementary), Python Refactoring (Intermediate), Gradio (Elementary), PyTorch Lightning (Elementary)

Mentors

Samet Akcay, Dick Ameln

Size of project

175 or 350 hours (depending on working on the optional step)

Difficulty

Medium/Hard (depending on choosing the optional step)

[sahilpmehra]

Hi @samet-akcay,

I would be very interested in contributing to this project idea for GSoC'23. I am almost done with my prerequisite task and would like to get started with any initial tasks/issues that I could solve related to this project. Could you help me with the same?

Thanks!

[samet-akcay]

Hi @sahilpmehra, thank you for your interest in this project! We are currently tagging some of the anomalib issues as Good first issue so the applicants can work on. I'll provide an update here once it is done. Cheers!

[sahilpmehra]

Awesome! Thanks for the heads up @samet-akcay. Looking forward to it.