Ex-MyFitnessPal CV engineer surveys industry: what's deployed, and what still hurts
kells1986 · reddit · 2026-09-26
A veteran computer vision engineer—who previously built the food recognition model for MyFitnessPal—runs an industry survey on what's actually deployed in production today: edge models, self-hosted models, or API calls, and what drives the choice.
He's most interested in unsolved pain points: problems that cost significant time, blocked delivery, or demanded awkward workarounds, hoping to find opportunities to build useful tooling rather than guessing at needs. He also asks which communities and forums CV practitioners frequent.
For developers considering tooling around computer vision, the thread is a first-hand demand signal.
More from coding & agent
- Qwen3.8-Omni-Flash: natively multimodal agent model with 1M-token context — dair_ai · 2026-09-26
- Zero-dependency LLM SDK lm15 1.0 hits PyPI, supports 10+ providers and already used by dspy — lateinteraction · 2026-09-26
- Dev builds Mac voice-to-computer-use app with no LLM, using native speech-to-text plus JEV — niloofar_mire · 2026-09-26
- PR Lens turns every pull request into animated architecture diagrams, powering 2,500 PRs a month — nevrekaraishwa2 · 2026-09-26
- LangChain's founder: Jev + LangGraph is the combo for modeling agents as complex systems — hwchase17 · 2026-09-26
- Want Opus as main agent in Codex desktop? One dev is stuck between toolchains — kevinkern · 2026-09-26