How Well Do Agricultural Self-Supervised Features Transfer? A Few-Shot Comparison of LeafVision and ImageNet-DINO
This manuscript reports an exploratory comparison of two publicly released self-supervised ResNet-50 encoders for few-shot plant disease recognition. The experiments have not been independently reproduced. PlantWild results are validation-only, and the study has limitations concerning pretraining-data overlap, test-set exposure, and reproducibility. This work has not undergone peer review.
Authors
- Abdussattar Sagyngali
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-15
- DOI
- https://doi.org/10.5281/zenodo.24246772
- Primary Topic
- Smart Agriculture and AI
- Type
- preprint