MosquitoScope — ConvNeXt V2 Tiny (41 classes)
Production checkpoint for the MosquitoScope web demo: 41 mosquito species, hybrid class-weighted training (v2_enhanced_memsafe).
Metrics
| Metric | Value |
|---|---|
| Test accuracy (TTA) | 74.0% |
| CV mean accuracy | 73.9% |
| Architecture | convnextv2_tiny @ 288px |
| Classes | 41 |
| Train / test | 1228 / 308 |
Usage
import torch
import timm
checkpoint = torch.load("mosquito_convnextv2_tiny_final.pth", map_location="cpu", weights_only=False)
model = timm.create_model(checkpoint["model_name"], pretrained=False, num_classes=len(checkpoint["classes"]))
model.load_state_dict(checkpoint["model_state_dict"])
model.eval()
Bundled with the MosquitoScope Flask app; Render pulls this file at deploy time via HF_MODEL_REPO.
- Downloads last month
- -