A web application leveraging a flask backend to detect, isolate, and store license plates from video. We used Google Vision API for OCR and database for storing best results. Automatic License Plate Recognition systems aim to provide a solution for detecting, localizing, and recognizing license plate characters from vehicles appearing in video frames. However, deploying such systems in the real world requires real-time performance in low-resource environments. In our project, we proposed a two-stage detection pipeline paired with Vision API that provides real-time inference speed along with consistently accurate detection and recognition performance. We used a haar-cascade classifier as a filter on top of our backbone MobileNet SSDv2 detection model. This reduces inference time by only focusing on high confidence detections and using them for recognition. We also imposed a temporal frame separation strategy to distinguish between multiple vehicle license plates in the same clip. We achieved a reasonable detection and recognition performance (82.7% detection rate, and 60.8% OCR F1) with real-time processing speed (27 fps).

We applied Conditional Generative Adversarial Networks (C-GANs) specifically tailored for Image-to-Image translation, using the Pix2Pix framework. For the generator, we implemented a sophisticated DoubleU-Net architecture, which is known for its powerful image segmentation capabilities. To enhance the robustness of our model, we employed a patchGAN as the critic. This architecture was particularly effective in segmenting skin lesions, a critical task in medical image analysis. Our novel approach yielded impressive results. We achieved a Dice Similarity Coefficient (DSC) of 89.7%, which is a significant improvement over the performance of a standalone DoubleU-Net model. The inclusion of the patchGAN critic played a crucial role in this enhancement, providing adversarial feedback that refined the segmentation accuracy. This advancement demonstrates the potential of combining Conditional GANs with advanced network architectures to improve the precision of medical image segmentation tasks.

In this project, we used approximate Q-Learning to teach a cat agent the game of jumping across buildings. We created the game environment and animations from scratch. At any given state, the actions available to the cat agent are either JUMP or STAY. Depending on its actions, the cat receives rewards that are stored inside a Q Table. By using the classic Bellman equations, we implemented an exploration vs exploitation strategy to maximize the cat's expected utility at each step. After about 1 minute and 30 seconds of playing, the cat figures out the mechanics of the game and learns to jump just at the right time to avoid falling from the buildings. Also during the learning stage, we made sure to put a pillow on the ground to ensure a safe landing for the cat. So we can safely say that no cats were harmed while making this project!

We developed a comprehensive full-stack website utilizing a combination of HTML, CSS, and Bootstrap for the front-end design, and JavaScript along with jQuery for dynamic interactions and client-side scripting. The back-end was powered by Django, a high-level Python web framework known for its clean design and ease of use. For data management, the website was seamlessly integrated with an Oracle Database, ensuring robust and reliable data storage and retrieval. The website featured user authentication modules, including login and sign-up functionalities, complete with thorough validation processes to ensure data integrity and security. Additionally, the website included advanced search and filtering algorithms, enabling users to dynamically search through content and apply various filters to refine their search results. This combination of technologies and features provided a responsive, user-friendly, and secure platform.