One of the toughest parts of our project was creating and training our own dataset from scratch. Since there wasn’t any ready-made dataset that exactly fit our requirements for sign language recognition, we had to become our own data collectors. We spent a lot of time capturing sign gestures manually, making sure every frame was clear, consistent, and usable. It sounded simple at first — but trust us, getting good quality data in different lighting, angles, and hand positions was way harder than expected. On top of that, cleaning the data and preparing it for training took a lot of trial and error. Giving our best, the model atleast worked finally.