Skip to content

README.md

Latest commit

 

History

History
73 lines (51 loc) · 2.65 KB

README.md

File metadata and controls

73 lines (51 loc) · 2.65 KB

🌅 Should I Go See the Sunset?

Ever stared at the horizon two hours before sunset, wondering if you’ll be blessed by a glorious sky… or just catch smoggy orange mush? Worry no more! This highly scientific, probably overengineered project evaluates sunset potential using images taken at T−2h and T−1h. It predicts—on a 1-5 scale—whether it’s worth dropping everything and sprinting to your nearest west-facing beach.

What Is This?

A PyTorch-based dual-image classifier that consumes sunset images taken 2 hours and 1 hour before sunset, and predicts a sunset "glory score" (1-5).

Yes, it's trained on real data. Yes, there's a GUI. No, we don't guarantee enlightenment, but you might achieve it anyway.

Main Scripts

File Purpose
create_sunset_model.py Builds, trains, evaluates, and predicts sunset quality using a model.
order_images.py Organizes raw images into a structured JSON file for training.
sunset_images_grouped.json Output file that links image pairs and their manually assigned scores.
img/ob_hotel/ Directory where sunset image files live (not included here).
model/ Folder for saving/loading trained models.

Model Architecture

  • Dual ResNet-18 backbones (shared or separate) 🤖
  • Processes -2h and -1h images separately → feature concat → classification
  • Outputs a score from 1 (💩) to 5 (🔥) based on sunset beauty

GUI Mode (Because Buttons > CLI)

Modes:

  • Train
    • Train the model with one click
  • Predict
    • Choose two images: one from -2h, one from -1h
    • Press "Predict Score"

🧪 Training

Make sure sunset_images_grouped.json is populated with:

  • At least one pair of images for each date (type: "-2h" and "-1h")
  • A score between 1 and 5

Training parameters:

  • ResNet18 ×2
  • 10 epochs
  • 64×64 images
  • Batch size: 32
  • Optimizer: Adam
  • Loss: CrossEntropy

🧙 Data Preparation

Run order_images.py to generate/update your training JSON file based on existing .jpg images named like:

25_07_24--18_45_00.jpg

🧙‍♂️ Why?

Fun excuse to play with:

  • Paired-image classification
  • PyTorch
  • PyQt6
  • Sunset FOMO

TODO

  • Automatically grab data via some google cloud or PA or something
  • Host somewhere and have it send a message that the sunset will be fire

License

MIT. Use it freely. Don’t sue if you miss the best sunset of your life.