Topic
Computer Vision
Exploring how neural networks learn visual representations for image classification, landmark detection, object detection, segmentation, and large-scale image recognition.
Computer Vision explores how machines can learn meaningful representations from images and use those representations to recognize objects, classify scenes, locate important features, and understand visual structure.
My exploration started with relatively constrained image-classification problems such as MNIST and CIFAR-10 and gradually expanded toward facial landmark detection, object detection, background removal, and large-scale image classification with ResNet.
A recurring question throughout this journey has been: what makes a neural network learn useful visual features?
This led me to experiment with convolutional neural networks, receptive fields, normalization, augmentation, depthwise separable convolutions, dilated convolutions, residual connections, object detection, segmentation, and model optimization.
The projects in this topic helped me move from simply training image classifiers toward understanding how modern computer vision systems are designed and how architectural decisions influence accuracy, efficiency, and generalization.