FrothSeg: a real model, and the honest fact that the data does not exist
Published:
FrothSeg: a real model, and the honest fact that the data does not exist
FrothSeg segments flotation froth into individual bubbles, entirely client-side via ONNX, and compares a ladder of seven classical methods against a real published model, LamellaStar (a four-head net shipped as a three-seed logit-mean ensemble, with SAM2 and Cellpose-SAM as offline teachers distilled to a compact browser head).
The honest core is the data situation, and I want it stated plainly rather than hidden behind a synthetic score:
The field's real blocker:
there is no public labelled real froth dataset. The search is recorded as a null (Roboflow dropped).
Froth cases are synthetic and labelled as such. BBBC038 (64 real dense-touching images, CC0) is an adjacent-domain transfer probe that does NOT clear the froth blocker, and the transfer study refuted its own hypothesis: the ranking is generator-specific.
there is no public labelled real froth dataset. The search is recorded as a null (Roboflow dropped).
Froth cases are synthetic and labelled as such. BBBC038 (64 real dense-touching images, CC0) is an adjacent-domain transfer probe that does NOT clear the froth blocker, and the transfer study refuted its own hypothesis: the ranking is generator-specific.
A real model, a real classical ladder, real transfer numbers, and a loudly stated limit. That published null, not a flattering synthetic accuracy, is the deliverable. A research lab, not a plant tool, part of the Faena hub. Live · source.