IdeaWell

Project ideas from Hacker News discussions.

Identifying fake cosmetics using AI

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

Three prevalent themes in the discussion

  1. Simple, low‑tech solutions can be surprisingly effective
  2. “This guy's lab does some really cool stuff using really basic ideas” – olwmc

  3. “Thanks! I'm putting 'doing cool stuff with really basic ideas' in my bio… :)” – wgrover

  4. AI shows promise for detecting counterfeits but faces scalability and adversarial challenges

  5. “Even if these techniques did not produce false positives … I suspect that they would not scale well. If counterfeiters realized they were losing a lot of sales … they will just use AI to catch the errors themselves.” – fwipsy
  6. “Regarding scaling … I'm hopeful that future AI models … might be able to identify more subtle variations …” – wgrover

  7. “This is genius – I’ve had good look using AI to identify fakes … It’s surprisingly good, and works fast enough to be usable in a fast paced live auction environment.” – trollbridge

  8. Counterfeiters often avoid exact copies due to practical constraints (OCR, source images, cost) and may adapt using AI

  9. “It's baffling that the counterfeiters don't use the exact same design for the box. At least using the same text. Almost seems harder to make text thats sort of close but not just a straight copy.” – qgin
  10. “Because they don’t have the source images. If they just used scans, the text would be fuzzy when reprinted. So they are converting it back to text in editing so it can print perfectly. Aside from some mistakes in the OCR.” – Gigachad
  11. “Any competent graphic designer can do this. I think it’s a lack of effort because… they don’t need to spend the extra $100.” – trollbridge

🚀 Project Ideas

Generating project ideas…

PillScan AI: Open-source Counterfeit Pill Detector

Summary

  • Detects counterfeit pharmaceutical pills by analyzing surface texture, injection molding patterns, and OCR-verified text.
  • Provides a pretrained model and simple API for developers to embed authenticity checks in apps or workflows.

Details

Key Value
Target Audience Pharma manufacturers, regulators, hobbyist collectors, online pharmacies
Core Feature Upload a pill image → receive authenticity score with explainable heat‑map
Tech Stack Python, PyTorch, OpenCV, FastAPI, Docker, optional TensorFlow Lite for edge
Difficulty Medium
Monetization Revenue-ready: SaaS tiered pricing per API call (free tier for low volume)

Notes

  • HN users wgrover and trollbridge highlighted the need for AI that spots subtle variations like molding patterns and works fast enough for live verification.
  • Provides a concrete tool that addresses the scaling concerns raised by fwipsy and wgrover, offering a community‑driven model that can be updated as counterfeiters evolve.

CandyCode Verifier: Mobile App for Micro‑Pattern Authenticity

Summary

  • Lets manufacturers generate unique, edible “candycode” micro‑patterns on product packaging and consumers verify authenticity via a smartphone scan.
  • Combines cryptographic hashing of the pattern with on‑device image matching to prevent replication.

Details

Key Value
Target Audience Product manufacturers, consumers, retailers, brand protection agencies
Core Feature Generate candycode, store its hash on‑chain/IPFS, verify by matching phone‑captured pattern against stored reference
Tech Stack React Native (expo), Node.js/Express, Solidity (or IPFS), TensorFlow Lite for pattern matching, QR‑code fallback
Difficulty High
Monetization Hobby (open‑source core, optional paid hosting for enterprise hash storage)

Notes

  • cmrx64 praised candycodes as “brilliant” for protecting intimate product chains; hn_throwaway_99 expressed cautious optimism about AI’s role in consumer protection.
  • A verifier app turns the theoretical candycode concept into a practical, consumer‑facing tool that HN discussants could readily adopt and improve.

AntiFake Hub: Collaborative Model Zoo & Dataset Marketplace for Counterfeit Detection

Summary

  • Central repository for sharing labeled datasets, pretrained models, and benchmark results focused on anti‑counterfeit AI (pills, packaging, surfaces).
  • Enables versioned model sharing, automated re‑training pipelines, and community discussion to stay ahead of evolving counterfeits.

Details

Key Value
Target Audience AI researchers, security startups, brand protection teams, academic labs
Core Feature Upload/download datasets/models, run standardized benchmarks, comment and fork improvements
Tech Stack GitLab/GitHub + DVC for data versioning, FastAPI backend, React frontend, CI/CD for automated testing
Difficulty Medium
Monetization Revenue-ready: Subscription for private repos + premium GPU compute minutes (free public tier)

Notes

  • trollbridge wanted to learn more about AI for reproductions; fwipsy warned that counterfeiters will adapt, underscoring the need for a shared knowledge base.
  • By providing a living hub of models and data, AntiFake Hub directly addresses the scaling and adaptation concerns raised in the thread, giving HN users a concrete place to collaborate and improve detection techniques.

Read Later