Home/Open Crowdfunding/Fridge Freshness Vision
🧊

Fridge Freshness Vision · A fridge that knows how fresh your food is

A time-series visual freshness system for apples, eggs, leafy greens and dairy: ESP32-S3 edge nodes + cloud vision LLMs (dual-vendor primary/backup) + a time-series scoring engine outputting a 0–100 freshness score and four states. Solution partnership for fridge OEMs and integrators, open for crowdfunding participation.

  • 🎯Crowdfunding goal from ¥20,000
  • 💰Solution partnership for OEMs & integrators
Current progress · updated 2026-0960%
🔧 Phase-1 in development · Cloud pipeline shaped
CrowdfundingOpen · goal from ¥20,000
Status● Live — accepting backers
📞 19040667970 Back this project
Highlights

What it brings you

🥬

0–100 freshness score

Baseline vs daily photos compared over time; four states (fresh / slight wilt / partial spoilage / spoiled) + estimated remaining shelf days.

🔀

Dual-vendor vision LLMs

Qwen-VL primary + Doubao vision backup with automatic failover; Wenxin VL / GLM-4V on the candidate list.

📷

Harsh imaging handled

Optimized for low light, condensation fog and packaging glare: fill-light 3-frame burst + anti-fog lens.

📡

ESP32-S3 edge nodes

Door-magnet / scheduled dual triggers, offline caching with resync, deep-sleep power saving.

🧮

Time-series scoring

EMA smoothing, rebound clamping and jump re-checks; strong output validation (≥99% format compliance).

💰

Cost under control

Monthly budget circuit-breaker and full call accounting; hard cases auto-saved as training material for the phase-2 in-house model.

Progress · 2026-09

Transparent R&D progress

The percentage is assessed from completed milestones across the whole development plan, and updated as development moves forward.

✅ Completed

  • Cloud recognition service complete (FastAPI + admin console, runnable, mock demo supported)
  • 86 automated test cases (cloud 60 + legacy gateway 25 + edge policy 1)
  • Dual-vendor vision-LLM automatic failover pipeline working
  • ESP32-S3 edge-node firmware complete (ESP-IDF 5.x)
  • Phase-2 groundwork: RK3588S local inference + YOLOv8 in-house training pipeline reserved

🚧 In progress / next

  • Hardware prototype integration and real-scenario evaluation set collection
  • Prompt and scoring-strategy tuning on real data
  • Phase-2 on-device in-house model (RK3588S + YOLOv8) kickoff
Technology

Key technology

CoverageCore 4 categories: apples / eggs / leafy greens / dairy; plus 20+ common food types
Edge nodeESP32-S3-N16R8 + OV5640 5MP anti-fog lens + fill light
CloudPython 3.10 + FastAPI + SQLite; dual-vendor vision-LLM API primary/backup
Scoring0–100 score + 4 states + estimated shelf days; EMA smoothing / rebound clamp / jump re-check
ModelSubsystem solution partnership for fridge OEMs / integrators; private deployment supported
RoadmapPhase 2 switches to on-device in-house model (RK3588S 6-TOPS + YOLOv8) with zero business-system changes
How to join

Back the Fridge Freshness Vision crowdfunding

🚀

Early access · Founder pricing

Crowdfunding backers get priority access to engineering samples and founder pricing, plus first updates on mass-production schedules.

🤝

Co-development · Custom

Enterprises can co-develop on any active project: custom sensors, enclosures, apps and cloud platforms — sharing the R&D outcome.

📈

Channel · Industrial partners

Distributors, system integrators and industrial investors are invited to participate via crowdfunding or strategic partnership.

Who is this project for?
Primarily fridge / freezer OEMs and smart-kitchen or cold-chain integrators: join solution validation and pilot deployments. Individuals and investors interested in the project itself can also participate via open crowdfunding (goal from ¥20,000).
How is recognition accuracy ensured?
Evaluation rules are solidified into prompt assets with strong output validation (≥99% format compliance), continuously tuned against a ~900-image evaluation set. Hard cases are auto-collected as training material for the phase-2 on-device model, so capability keeps improving.
Risks & Challenges

We put risks on the table

Crowdfunding means sharing development uncertainty with us — please read these before backing.

  • Field-tuning risk: long-cycle accuracy in real fridges (fog, glare, occlusion) requires prototype integration; evaluation data will be published on the project page.
  • LLM service dependence: cloud recognition uses third-party vision APIs; dual-vendor failover mitigates outages and the edge version removes dependence entirely.
  • Category coverage: 4 core categories are finely calibrated; accuracy is not promised for items outside the supported list.
  • Delivery timeline: the 9-week standard schedule may extend depending on prototype and data availability; milestone-based and transparent.

Got an idea? A machine to make smart?

From a brand-new AI device to retrofitting existing equipment — talk directly with our engineers. Bilingual service in English & Chinese.

📞 Phone / WeChat: 19040667970
✉️ buy@allinai5.com Mon–Fri 9:00 – 19:00 (UTC+8) · Shenzhen, China · Serving worldwide