ATLien415

<span class="link_user">@Ivy_Vivosun</span>, Sounds rigorous. :ok_hand: Mmmm boundary value cases and real-world variability make a lot of sense. When I was referring to available models, I mean more of academic machine learning models specifically for image classification that are open-source. I would also be hesitant to grab a vendor model off the shelf in the current environment as well. Will Vivosun be keeping the community updated on this AI GrowCam?

<span class="link_user">@Ivy_Vivosun</span>, I must say this is much more information than I expected, and I genuinely appreciate it. My first reaction to this is curiosity. Managing condensation is managing dew point, so the same as logic as the Cannatrol. This is expected as this is the same logic for antiques/books/meats/cheeses/and more. I can see how from an engineering standpoint a first pass would be where the raw power is (compressors). Does this not have some impact on the local environment? From my understanding, and this is my words of whitepapers that have been reproduced, is that trichome cuticles measure a couple molecules thick and that things on the order of vapor pressure gradients (essentially invisible waves caused by things like compressors) were enough to rupture heads...especially in dry/cure. Ruptured heads leak, and by that mechanism you lose secondary metabolite mass. The Cannatrol team has had this finding of increased trichome integrity both by reproduced manual gridding/counting of heads and a (novel?) chromatography method. Have yall confirmed that you are on the same order increase as Cannatrol? Granted, a TEC will never last as long as a compressor device...a TEC doesn&#039;t mechanically change the environment of the flower when I think about how the device functions. I guess that is my original concern with the design and something folks like myself have questioned since the announcement until now.

<span class="link_user">@ATLien415</span>, On the VCure and compressor design: This is a compressor-driven system. The design choice over semiconductor cooling comes down to power, precision, and longevity — variable-speed compressor technology gives us tighter environmental control with better durability over time, paired with our own algorithm managing internal condensation to maintain stable conditions. On trichome integrity specifically: The mechanism isn&#039;t about any single feature. It&#039;s about eliminating the environmental volatility that causes damage in the first place. Hang drying and traditional curing expose material to fluctuating temp and humidity, which causes trichome stalks to become brittle and increases fracture risk significantly. By holding both tight throughout the entire post-harvest window, along with consistent gentle airflow, we&#039;re keeping the gland structure stable without physical contact at any stage.

<span class="link_user">@ATLien415</span>, Great questions! The level of technical depth here is exactly the kind of feedback we find most useful, so thank you for taking the time. On the AI GrowCam data pipeline: To clarify, user submissions are not directly used as training data. We operate a multi-stage data pipeline: User submissions (images + metadata) Automated quality filtering Model-assisted pre-labeling Human validation (internal or expert-reviewed) Consensus-based verification This ensures that only high-confidence, structured data enters the training set. Regarding crowdsourcing: the goal is not to replace expert knowledge, but to expand coverage of real-world variability — including lighting conditions, plant stages, and rare edge cases that are difficult to capture in controlled datasets. We also explicitly model label uncertainty using techniques such as consensus labeling and confidence-weighted training, rather than assuming a single &quot;ground truth&quot; per sample. Finally, general-purpose models are limited in this domain due to lack of specialized data and fine-grained visual understanding. Our focus is on building a domain-specific system with a continuous data feedback loop, rather than relying solely on off-the-shelf models.

<span class="link_user">@Ivy_Vivosun</span> , To be clear...the submissions for this competition will be used as training data for your AI resource? What is that going to mean? Who is labelling this set/what is the purpose of the user submitted label? There are some amazing free, locally ran image processing models that are ubiquitous in industrial applications. Something like that on board with a context layer to integrate into some type of bounded LLM on your hubs would be interesting. The newest folks in this space are taking non-visible light spectrum for more fun. My questions are this... Why crowdsource this data? Isn&#039;t the veracity of this data the issue in the first place.......or else ChatGPT would be diagnosing plants accurately. The simple fact is that these online spaces like Reddit and whatnot are going to give you a picture of N toxicity and you&#039;ll have 5 distinct answers and people ready to draw weapons over &quot;their&quot; response, only one of which was correct. Data labelling for anything of pedigree which isn&#039;t just common knowledge...it is a job description, and it pays particularly well in most circumstances. It also has an extreme bar to clear for both accuracy and precision. I just have questions. Like for example, when yall crowdfunded your Vcure. You made a whole lot of marketing noise about an &quot;inverted compressor&quot; which is fine. Compressor-driven devices run on exploiting work cycles and you can invert that, which is a common thing. I&#039;m still not sure about the &quot;inverted compressor&quot; itself as to the device but from my understanding it is a compressor-driven device. Why make that design choice? From my viewpoint, yall dropped a crowdfunded cabinet that looks like a Cannatrol but ignores the entire point of the design. Dew point modulation for permanently mold theory, aka the orthogonal feedback loop is just half of the story. The other half was and is trichome integrity. How are you maintaining trichome integrity with a compressor?