Week 2: June 09 ~ June 15 | Md. Shariar Kabir

Preliminaries

In the weekly meeting held on 06 June, we discussed the previous work and outlined the project's current focus, which is to extend the capabilities of the RoboticsAcademy platform by integrating deep learning models in ONNX format into both existing and new exercises. The goal is to enable users to upload their own models and run inference within the RADI (RoboticsAcademy Docker Image) environment. Initial work includes exploring the architecture of the RoboticsAcademy Docker image, creating Human Detection and Digit Classification exercises, and integrating a file upload mechanism for ONNX models. This week also involves studying PyTorch to better understand deep learning workflows and ensure compatibility during ONNX model conversion and deployment.

Objectives

Convert semantic segmentation from labels to RGB images
Create Human Detection and Digit Classification exercises
Implement support for user-uploaded ONNX models
Beta test the exercises to ensure functionality and usability
Update local exercise database
Study PyTorch and ONNX

Bug Fixing

In production mode, I discovered that the education link in the Computer Vision exercise was incorrect. This issue caused navigation errors for users attempting to access related resources or documentation. I have identified the source of the incorrect link and am currently working on updating it to ensure proper functionality in the production environment.

Issue: #3106

Pull Request: #3113

Execution

I created two branches, one for human detection and another for the digit classification exercise. This branch that contains both exercises. I also updated the local database. This involved updating the exercise metadata and ensuring that both new exercises are properly registered and accessible through the RoboticsAcademy interface. I also verified that the ONNX model upload and inference features work seamlessly with the updated database, allowing users to test their own models within the new exercises.