Thanks for
Data provided by TerraWave Co., Ltd.
FROM IMAGE TO HARVEST · 2026
Following one plant,
from image to harvest decision.
Follow photographs and observations of the same bed over time, from raw images to growth measurements and harvest decisions.
01 — OBSERVATION RECORD↓ Explore the data below
SOURCE NOTE / 01
Every number is linked to a source image, capture date, physical slot, and run result. Gray trend values are for display only; original measurements and model inputs are retained.
01 / OBSERVATION EXPLORER
Explore Bed45 over time
Select a slot in a bed image to inspect its result and review evidence for that date.
—
SOURCE IMAGE
Original image
—
The boxes on the image indicate physical slot locations. They are separate from lettuce detection boxes and masks.
SELECTED PLANT
Selected slot
Select a slot to enlarge its location in the source image.
—
MODEL OUTPUT
Harvest decision
Predicted fresh weight——
Predicted heading/growth stage (0–4)—
No separate binary heading classifier is used.
Harvest decision—
Estimated harvest window——
—
02 / GROWTH TIMELINE
Trends as lines, evidence as points
These are observations of the same physical slot. Plant identity over time depends on review status.
MASK-BASED AREA
Top-view area
Estimated cm² converted from mask pixels using image scale.
Original mask area estimateDisplay trendReviewed / flagged point
Only observations from the same camera are connected. The gray trend is for display, not a model input.
CURRENT OBSERVATION
Selected-date metrics
GROWTH QUALITY CONTROL
Growth QC
ENVIRONMENT CONTEXT
Growth environment
Daily observations from shared facility sensors; these are not Bed45-specific sensors.
Same-day shared facility observations. The time boundary for model inputs is reviewed separately.
03 / HOW IT WORKS
From source image to decision
See how the selected date and slot pass through each step using actual evidence.
——
—
View full image ↗—
—
View model input ↗View mask ↗—
Growth and environment—
—
Stage, weight, harvest—
Checking run lineage
Source and bed alignment
Identify capture time and camera, then align Bed45's physical coordinates.
Slot, crop, mask
Crop each slot and review fragments and clipped boundaries.
Growth and environmental context
Evaluate area, front-view pairing, time series, and environment available before capture.
Stage, weight, harvest
Pass verified inputs to the models and present predictions with uncertainty.
OPEN AUDIT TRAIL
Predictions and evidence,
together.
Predictions, areas, and QC states link to Bed45 audit data. Unavailable results and review cases remain visible.
Data provided by TerraWave Co., Ltd.Research direction, analysis, and result verification: YeonjinSite planning and review: Yeonjin · Implementation assisted by ChatGPT
For Haengbok, my beloved cat who stayed beside me through this research. Thank you, and I love you.
Measured harvest records remain in the private audit workbook.
Loading run information