• Kshitij Yerande - 40194579
• Siddhartha Jha - 40201472

The goal of the project is to analyze the performance of the various machine learning approaches to solve image classification problem of classifying sea animals into 9 categories Corals, Crabs, Penguin, Sea Urchins, Seahorse, Seal, Sharks, Starfish,Turtle_Tortoise.

The dataset is retrieved from kaggle.

Link: https://www.kaggle.com/datasets/vencerlanz09/sea-animals-image-dataste

1. Accuracy: It is the most intuitive performance measure, and it is simply the ratio of correctly predicted observation to the total observations.
2. Precision: It is the ratio of the correctly predicted positive observations.
3. Recall: It is the ratio of correctly predicted positive observations to all observations in actual class.
4. F-score: It is the weighted average of precision and recall.

• Custom-CNN model: CNN architecture desined from scratch
• ResNet50 (Pre-trained) : Pretrained weights of ImageNet dataset used to train last layer on our dataset.
• SVM with HOG features(histogram of gradients) : HOG feature extracted and trained using SVM
• SVM with SIFT features: SIFT features extracted and trained using SVM

• Sea-Animals-Classification
  • data/archive
  • models/
  • metric/
  • Sea Animals Classification.ipynb

Project Setup

1. Install libraries:
   • Pytorch
   • Numpy
   • matplotlib
   • Scipy
   • Scikit-learn
   • OpenCV
   • pandas
   • tqdm
   • torchviz
   • skimage
2. Download the data from the kaggle repository. Place all the classes in the following format.
   • data
     • /archive
       • /corals
         • 1.png
       • /seahorse
         • 2.png
3. Create the empty folders as per directory structure
4. Run the notebook