HANDS-ON-LAB

Water Bottle Identification Project

Problem Statement

In this exercise you are going to build an image segmentation model that identifies your water bottle from an image.

Tasks

  1. Create your own water bottles dataset:

  • Take your water bottle, keep it in your living room, bedroom, kitchen, etc, and take a few pictures with the whole surrounding. (try to take 30-40 pics)

  • Along with it also google the images and collect a few from it (try to get 20-30 pics)

  • Create 3 folders - Train (40 images), Validation (20 images), Test  (10 images)

  1. Use OpenCV library to resize all the images to 256 X 256 pixels (Link)

  2. Use image augmentation technique to manually increase the train & validation dataset (link)

  3. Use the APEER Annotation tool to annotate the images collected.

  4. Upload your dataset - images + annotated files to the s3 bucket.

  5. Build the Unet model to identify & segment eye-glasses from a given image. (follow the same process as described in the project video)

 

Explore the process of identifying and segmenting water bottles from images.

FAQs

Q1. What techniques can I use to increase my dataset size?

Utilize image augmentation techniques to generate variations of the existing images.

 

Q2. Which model architecture should I use for water bottle segmentation?

Build a Unet model, a popular architecture for semantic segmentation tasks like object identification and segmentation.