Polaris

• 

  Polaris HomePage

• 

  Parking Analysis

• 

  Polaris Model Architecture

• 

  GrowthFactor Parking MVP System Design

Inspiration

We are Polaris and we built a north star that guides retailers to their perfect location, based on insights about parking.

Our project solves a real problem that retailers face: locations that look perfect on paper but might lead to millions/years in potential loss of foot traffic due to hidden parking dynamics.

What it does

Polaris is a scalable ML platform that quantifies parking inventory from space. Drop a pin anywhere, and in under 2 seconds, we parse satellite imagery to deliver stall counts, confidence intervals, and a 0–100 Polaris Score for accessibility.

How we built it

We tried 3 approaches, all 3 have their pros and cons:

1. Segmentation Transformer model: creates highly accurate masks of parking lots, but struggles to differentiate between individual stalls (computationally expensive)

2. YOLO model: creates bounding boxes around each stall and is able to detect cars, but is not very precise (fast)

3. Geometric Heuristics: uses OSM + SegFormer boundaries to estimate stall counts based on lot geometry and optimal parking angles, but struggles with irregular shaped lots (near instant, highly accurate)

Challenges we ran into

1. Training the SegFormer model was both time consuming and expensive
2. Handling edge cases such as multi-story parking lots and street parking
3. Adapting real world insights such as optimal parking angles to our model

What we learned

Raj taught us a lot about how to think about how this project fits into the entire ecosystem of Growthfactor.

Built With

• fastapi
• python
• react
• vultr