Task3: Build a Custom CNN Model for the Dataset

[saumyacoder1709]

Merry Christmas contributors!

Task Overview:

Participants are required to create a Convolutional Neural Network (CNN) model for the given dataset that achieves at least 40% accuracy on the test data. The model must follow the specified requirements to ensure consistency and quality.

You are given the Task3.ipynb. You can either copy it a the end of your task2 notebook OR Write the code specified in the notebook again. Show clear output of all the cells.

---

Requirements:

1. Directory Structure:

   • Ensure the directory structure adheres to what was outlined in Task 1.
   • File naming conventions must remain consistent with the guidelines in Task 1.

2. Data Handling:

   • Use ImageDataGenerator for preprocessing and augmentation.
   • The validation set can be created within the ImageDataGenerator itself. If there exists a separate directory of validation_set then it can be used for validation. No issues!
   • Apply necessary augmentation techniques to enhance model performance.
   • Ensure the dataset is balanced for better results.

3. Model Details:

   • The CNN model must be custom-built (no pre-trained models allowed).
   • Set verbose=1 in the model.fit() method to display accuracy, loss, validation accuracy, and validation loss during training.

4. Evaluation Metrics:

   • After training, print the following metrics on the test set:
     • Test Loss
     • Test Accuracy

5. Visualization:

   • Include the following plots in your notebook:
     • Accuracy vs Validation Accuracy
     • Loss vs Validation Loss

6. Submission Guidelines:

   • Provide the Colab Notebook link in a file named Task3_solutions/rollno.txt (e.g., iit2023098.txt).
   • Restrict access to the following email IDs only:
     • saumyagrawal1709@gmail.com
     • iit2023098@iiita.ac.in

---

Important Notes:

• Ensure the dataset directory structure is followed as outlined in Task 1.
• Balance the dataset for better accuracy if necessary.
• The final model must achieve at least 40% accuracy for it to be considered valid for merging.

---

Deliverables:

1. Colab Notebook:
   • The notebook must include the CNN model, preprocessing, augmentation, training logs, and visualizations.
2. rollno.txt File:
   • Include the restricted-access Colab link in the Task3_solutions/rollno.txt file.

---

- Is a large dataset helpful?