Ask HN: Seeking Solutions for Automating Aesthetic Photo Cropping with CV/AI

Innovating K-Pop Merchandise Production with AI Automation

As a backend developer, I am currently engineering a cutting-edge automation tool for the K-pop merchandise industry, focusing on photocards and postcards. My project involves:

• Tech Stack: Python 3.11, FastAPI, Libvips, and InsightFace.
• Workflow:
  1. Bulk Upload
  2. Landmark Extraction (InsightFace)
  3. Auto-cropping with predefined ratios
  4. Human-in-the-loop fine-tuning via Web UI.

Key Challenges:

• Mechanical Logic vs. Aesthetic Crop: Current centering logic can’t capture the dynamic poses of K-pop idols, leading to awkward crops.
• Reference-Based One-Shot Style Transfer: Clients often desire specific visual “feels” from guide images for thousands of photos.

Seeking Insights:

I’m looking for direction on enhancing aesthetic composition and one-shot composition transfer. What effective algorithms or methodologies exist?

Your expertise could revolutionize this project. Join the conversation and share your thoughts!

Source link