K's Tomato Trails 2026

Phase 1: seedling growth from day 1 to the lock.

This landing page keeps the project story intact while turning the tomato view into a Phase 1 comparison board. We are still in the seedling stage: every pot card leads with the last day of Phase 1, and a flip reveals the day-one frame so the change is easy to read at a glance.

Phase Window

23 days

Locked Pots

32

Strong Starts

22

Watchlist

10

Project frame

Tomato Trails is a citizen-science backyard trial asking which varieties can really hold up in Sausalito's cool, humid fog belt. The page below is intentionally narrow: it is the locked seedling story, not the whole season.

The strongest Phase 1 use cases are survival checks, reference-pot selection, and deciding which seedlings need closer human follow-up before the next stage.

Reference cohort

These are the clearest Phase 1 examples to reuse in future reporting and segmentation work. They combine strong establishment with at least usable CV framing.

Follow-up queue

The watchlist is where the trial gets decision-useful: modest growers, the single stalled pot, and any seedling that needs a tighter future capture or in-person check.

Variety outlooks for Sausalito fog

These cards preserve the varietal read alongside the Phase 1 signal. The outlook is intentionally directional until later phases tell us who can really set and ripen.

Showing 32 of 32 pots Front of every card = Last Day of Phase 1. Flip the image to revisit Day 1 of Phase 1.

What is reliable now

The manual Phase 1 triage is the canonical growth read for this page. It is strong enough for establishment buckets, watchlist selection, visible mold-risk cues, and picking reference pots for future modeling work.

  • Reliable now: strong vs modest vs stalled establishment
  • Reliable now: early variety signal and watchlist generation
  • Not yet reliable: exact cross-run size or biomass scoring from raw pixels

CV strategy for next phase

Because the camera angle and distance drift between the anchor captures, the next quantitative pass should align the pair first and only then measure plant material inside the target pot.

  • Register image pairs first to absorb framing drift before any growth math.
  • Keep the measurement inside the pot region so neighbor foliage does not fake growth.
  • Track canopy coverage, spill, plant count, and chlorosis together instead of relying on a single score.

Where a VLM helps

A vision-language model is useful here, but as a secondary layer. It should comment on legginess, empty pots, possible mold, or multi-stem competition after the numeric CV pipeline has done the measurement work.

  • Good use: draft qualitative notes and anomaly QA
  • Good use: compare the two images and describe visible progression
  • Not the primary tool for: numeric growth scoring or area estimation

Open the deeper pot-CV research page