Image Classification with Deep Learning
Uploaded by
tuan nguyenImage Classification with Deep Learning
Uploaded by
tuan nguyenFinal AI Project: Image Classification Program with Deep Learning
Objective:
Students will write an image classification program in Python using a self-designed deep learning network.
The program will classify images into eight categories with the highest possible accuracy.
Dataset:
• Use an image dataset from Kaggle (downloaded or trained on the platform).
• The dataset should include separate training and test sets with eight classes.
Network Design Requirements:
• Achieve the highest possible classification accuracy.
• Utilize advanced connections, structural topologies, and training techniques (recommended).
• Leverage built-in functions for program development.
• Maintain a total network parameter count under 300,000.
• Configure training parameters such as learning rate, batch size, and optimizer.
Evaluation:
• Run the code three times on the test set for 40 epochs each. Record accuracy at epoch 40 for each
run.
• Final accuracy is the average of the three runs.
Baseline Requirements (to Pass):
• Successful Program Execution: Program failure results in failing the class (no exceptions).
• Originality: Network architecture cannot be substantially similar to provided sample networks
(simply adjusting hyperparameters like adding layers or changing kernel size is insufficient).
• Performance: The network must meet both accuracy and complexity criteria concurrently:
o Accuracy: Above 85%
o Complexity: Total parameters less than 300,000
• Shortlisting: Only groups meeting both baseline requirements will be shortlisted. Shortlisted
groups will be ranked based on accuracy:
o Top 25% receive 7 points
o Next 25% receive 6 points
o Next 25% receive 5 points
o Remaining groups receive 5 points (maximum 7 points due to a 3-point deduction for the
technical report)
• Technical Report (3 points): A comprehensive report (in PDF format) is required, covering topics
such as:
• Deep learning background
• Relevant algorithms
• Your network architecture
• Design decisions
• Simulation results
• Discussion and insights
Submission:
• Submit both your Python code file (.py) and the technical report file (.pdf) to Google Classroom
before the deadline.
• Final Class Presentation: During the final week, be prepared to:
o Print your report (double-sided printing encouraged).
o Bring your laptop to re-run your code in front of the instructor in case of result
discrepancies.
Important Notes:
• Late submissions will result in a failing grade (no exceptions).
By completing this assignment, you will gain valuable experience in designing and implementing deep
learning models for image classification.