ATAIIR2604_NorthSea_multiclass

Fine-tuned ViT from the PISCO plankton imaging pipeline, run ATAIIR2604_NorthSea (ATAIR-BSH, North Sea, April 2026).

Classes (7): Appendicularia, Chaetognatha, Cnidaria, Copepoda, Ctenophora, gelatinous, pluteus

Important: this model is coupled to its preprocessing

Trained on crops produced with the v5a LUCYD deconvolution (lucyd-edof-plankton_v5a_20260807.pth) and neighbour-isolated crops (PISCO_ISOLATE_CROPS=1). Running it on crops from a different deconvolution model, or on non-isolated crops, is an out-of-domain shift and degrades the living/not-living split badly.

Preprocessing: resize longest edge to 224, centre-pad to 224x224 with white (255), normalise mean=std=0.5. No rotation at inference.

Use in the pipeline

process_pisco_profiles.py --dualvit-model ATAIIR2604_NorthSea
Downloads last month
9
Safetensors
Model size
85.8M params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support