Alohomora

This repository contains the implementation of the Alohomora vision pipeline, featuring a custom Pb-lite boundary detector using DoG, LM, and Gabor filter banks. It further benchmarks deep learning classifiers—including ResNet, ResNeXt, and DenseNet—on the CIFAR-10 dataset to evaluate performance trade-offs in robotic vision systems.

Course Homework for RBE549 - Computer Vision (Spring 2026)

Master of Science in Robotics Engineering at Worcester Polytechnic Institute

Prerequisites

• CUDA Toolkit + GPU drivers
• PyTorch
• NumPy
• Matplotlib
• OpenCV

Usage

Phase 1

1. Navigate to the Phase 1 directory:

   cd rnallaperumal_hw0/Phase1/Code

2. Run the wrapper script:

   python3 Wrapper.py

3. Outputs will be saved in the respective folders. Note: The paths in the Wrapper code are exact and need to be modified accordingly before executing the Wrapper.py file

Phase 2: Model Training and Testing

Training

1. Navigate to the Phase 2 directory:

   cd rnallaperumal_hw0/Phase2/Code

2. Run the training script:

   python3 Train.py 

   Note: DenseNet model is loaded by default. Other models are commented. Number of epochs by default is 50 and the minibatch is 1. Modify them as per requirement

Testing

1. To evaluate the model on the test set and generate a confusion matrix:

   python3 Test.py --ModelPath ../Checkpoints_dense/model.ckpt Train/Test

   Replace ../Checkpoints_dense/model.ckpt with the path to your saved model checkpoint. Train/Test to check the confusion matrix of either

References

• RBE549 Course Assignment