HANDS-ON TRAINING IN IMAGE PROCESSING

A five-day Online Hands-on Training session for image processing was organised for the students of BSc Data Science Students; The insightful session was taken by Mr Janarthnan M, who started with sharing the basics of Computer Vision and Image Processing to the advanced image processing concepts; the session was organised to help the students to gain more knowledge in image processing and to help them in completing the internship related to image processing.

Day- 1:Mr Janarthan talked about the basics of Image Processing and classic image processing algorithms. Analog and digital image processing are the two techniques used in image processing. The method for processing pictures, prints, and other tangible reproductions of images is known as analogue image processing. In contrast, digital image processing uses sophisticated algorithms to manipulate digital images to generate information. They were briefed about them with the help of a presentation. And introduced them in Colab. 

Day-2: Mr Janarthan talked about the package FiftyOne, the free software for creating superior datasets and computer vision models. Poor quality data is the biggest obstacle to the success of machine learning systems. Additionally, enhancing a model might be time-consuming and ineffective without the proper tools. By enabling you to quickly and accurately display datasets and comprehend models, FiftyOne accelerates your machine-learning operations. And explained them in python. It also explained how the dataset could be imported into colab.

Day-3:Mr Janartan started with an explanation of 3D object detection in colab. The majority of object detection research has been on 2D object detection. These are some of the models, RCNN, Fast RCNN, SSD, and Masked RCNN. Real-world environments contain 3D items. Because of this, 3D bounding boxes would be preferable to the often-utilized 2D detections for binding things seen in the actual world. We need 3D object detection to record the sizes, orientations, and locations of items in the real world. Since robotics see the environment similarly to humans, we could apply this 3D detection in practical applications like Augmented Reality (AR), self-driving automobiles, and robotics. Amazingly, Google has developed a model that perceives the environment and can identify three-dimensional things in the actual world. The objection is the name of this design. Explained the project implementation and enlightened for the dataset provided by Xoriant.

Day-4:Day with Mr Janarthan continued with the 3D object detection and solved the bug for the code of the students, The way object detection works in the video is quite similar to how it operates in photos. Such a tool would enable the computer to find, recognise, and categorise things visible in the supplied moving pictures. First things first, reference data must be fed into the machine, so for that, examples were provided. The fact that a video is a collection of individual pictures (or frames) is one obvious factor for the slight imbalance. However, because video processing introduces a new dimension to the problem, the time dimension of this description cannot fully capture what video processing is proper guidance for proceeding with the project.

Day-5:The insights on Pixel Level Scene Segmentation and the various types of image segmentation as there were a set of students provided with the pixel level scene segmentation which explained the different types, namely instance, semantic and panoptic, where we were guided with proceeding with the semantic segmentation. The project required proper explanations that were taught with the help of various examples through the coding part. Also, brief explanations were given, and the basics were explained using codes in colab. Further guided us about the execution part of the project.