0:00:00:00
Time until launch!
About Us
Kollec is a collection management platform developed as a final year Computer Science capstone project at McMaster University. Our mission is to bridge the gap between physical collectibles and digital organization using Computer Vision and Natural Language Processing.
Beyond organization, Kollec actively facilitates community engagement by intelligently matching users who possess viable, mutually beneficial trades.
Kollec can be found on GitHub:
Features
Card Identification
Using advanced machine learning techniques, Kollec can rapidly identify collectibles in real time using just your phone's camera. No more manual entry or searching through endless lists!
The Technology
Kollec first uses a custom trained YOLO segmentation machine learning model to locate any cards in view of your device's camera. We then use perceptual hashing to match the found cards against our database of known cards. Once a match is found, it is shown to you so you can quickly add the card to your digital collection.
Card Search
Kollec allows you to search your Pokémon collection using natural language. Whether you remember the exact name of the card or just a description of the Pokémon, our intelligent search understands the visual context of your cards to bring you the right results instantly.
The Technology
We utilize a CLIP (Contrastive Language-Image Pre-training) model to generate mathematical embeddings for every card image, which are stored and loaded at startup. When you enter a query, a quantized version of the model runs locally on your device to turn your text into a vector and compare it against our database. By calculating the shortest distance between these embeddings, the app identifies and displays the most relevant cards.
Trading Algorithm
Connect with local collectors to complete your set through our intelligent TradePost matching system. By syncing your location and wishlist, Kollec automatically pairs you with nearby trainers who have the specific cards you need and are looking for the ones you have.
The Technology
TradePost utilizes a location-based algorithm that converts user addresses into precise longitude and latitude coordinates to calculate real-time distances between collectors. The system filters the database for users with active "forTrade" flags and cross-references them against your specific WishlistEntry table to find mutual matches. By applying a customizable distance radius the algorithm ensures you only see relevant trade opportunities within a reachable proximity for safe, in-person exchanges.