Dataset
Three ways in, one format out (YOLO-pose: one class, four grid-corner keypoints). Merge what you have, look at the labels, then train.
datasets you build here are filed under this
project; change it any time on the Projects page
Synthetic
Render frames across modulations and grids, push them through the camera channel. Minutes of CPU, no phone needed.
≈ 0.15 s per frame per core. 300+60 ≈ 1–2 min on 4 vCPU.
Real captures
Photos from the capture station, labelled by the localiser (or the active detector). This is the data the sim gets wrong.
checking…
Need photos first: display on the laptop, camera on the phone.
Merge
Combine any datasets above into the one to train on. Filenames are prefixed, unlabelled images are dropped.
no datasets yet
A good default: all of synthetic + all of real.
Datasets
–loading…