Memory Maps
The Idea
This project explored the creation of user-driven spatial memory maps through an iterative development process. The initial concept focused on understanding how users manually organize and annotate memories spatially, using a basic interface without AI automation (V1). This provided foundational insights into natural user behaviors. Subsequent iterations aimed to modernize the technical foundation for scalability and improved user experience (V2), setting the stage for advanced features. The final stage integrated AI (LLaVA) to automate image descriptions, creating richer spatial narratives with less manual effort and demonstrating the potential for intelligent, context-aware memory mapping.
Version 1: https://calluxpore.github.io/Memorymap/
Version 2: https://calluxpore.github.io/Memory-Map-V2/
Version 3: Locally Hosted - Memory Map with AI on Vimeo
Development
- Initial Manual Prototype (V1): Built with HTML, JavaScript, and Leaflet, this version allowed users to click on a map to create markers, manually upload images, and type descriptive labels. Features included basic memory clustering and a sidebar for navigation. The focus was on observing organic user interaction patterns.
- Modernized Foundation (V2): Transitioning to React, Leaflet, Tailwind CSS, and TypeScript, this iteration introduced a component-based architecture for improved scalability and maintainability. Enhancements included better marker clustering, draggable markers, chronological connections between points, and a refined dark-themed UI, focusing on technical resilience and user experience improvements.
- AI Integration (Advanced Prototype): This version combined a Flask backend with the Leaflet front end and integrated OpenCV for real-time image capture linked to map markers. The core advancement was using LLaVA to automatically generate text captions for captured images. It retained features like chronological connections, a fullscreen gallery, sidebar navigation, and the refined dark UI, aiming to automate memory documentation.
Reflection
The iterative journey began by capturing genuine user behaviors in manual memory mapping (V1), providing a valuable baseline. The modernization phase (V2) yielded a more polished, scalable, and user-friendly platform with a robust codebase prepared for future enhancements. The integration of AI (Advanced Prototype) successfully demonstrated the feasibility of automating descriptions, significantly streamlining the memory documentation process and creating more coherent spatial narratives. While challenges like scalability for large datasets and the lack of AR overlays or on-device personalization remain, the project effectively proved the potential of combining interactive mapping with AI to create powerful tools for organizing and recalling visual memories spatially.
What Worked
- User Interaction & Automation: Started with intuitive, user-friendly manual actions (click, upload, label); progressed to improved usability via interactive markers and clustering; culminated in automated AI captioning that streamlined documentation.
- Organization & Navigation: Evolved from basic visual grouping to enhanced map clarity, aesthetics, and responsiveness; introduced chronological markers for a narrative flow, complemented by a fullscreen gallery and sidebar for better Browse.
- Technical Foundation: Gained crucial insights into user needs from the initial prototype; developed a maintainable, scalable architecture primed for future feature integration.
What Did Not Work
- Manual Effort: Initial reliance on user-generated text was laborious and remained a factor until AI integration partially addressed it.
- Scalability & Complexity: Early versions could become cumbersome with many markers; modernization increased setup complexity; AI integration introduced scalability concerns for large datasets and continuous processing.
- Missing Advanced Features: Lacked AI automation in early stages; AR overlays were not implemented; on-device personalization for adapting to individual user needs was absent in the AI version.