Blood Parasite Classifier — Howl Vision
EfficientNetV2-S for blood smear microscopy (8 classes: 6 parasites + 2 controls). Part of Howl Vision for the Gemma 4 Good Hackathon.
One Health: Leishmania, Plasmodium, Toxoplasma, Trypanosome are zoonotic.
Metrics (held-out test, n=3032)
| Metric | Value |
|---|---|
| Accuracy | 99.87% |
| F1-macro | 0.997 |
Details
- Architecture:
timm/tf_efficientnetv2_s.in21k_ft_in1k, num_classes=8 - Input: 384x384 RGB, ImageNet normalization
- Classes: Babesia, Leishmania, Leukocyte, Plasmodium, RBCs, Toxoplasma, Trichomonad, Trypanosome
- Caveat: Validated on standardized microscopy. Generalization to variable staining/magnification requires separate validation.
import timm, torch
model = timm.create_model("tf_efficientnetv2_s.in21k_ft_in1k", pretrained=False, num_classes=8)
ckpt = torch.load("vet_parasites.pt", map_location="cpu", weights_only=False)
model.load_state_dict(ckpt["model_state_dict"])
model.eval()