4 model checkpoints for the vit_huge_plus_patch16_dinov3 architecture trained using a 4-fold cross-validation strategy. These weights are specifically optimized for biomass prediction and achieved a 0.74 silver medal score in a CSIRO competition.
Use Cases
- Execute ensemble inference for biomass prediction by loading the 4-fold vit_huge_plus_patch16_dinov3 weights
- Transfer learn for other vegetation-related tasks using the pre-trained weights as a specialized backbone
- Benchmark new biomass estimation models against the 0.74 silver medal performance baseline provided by these checkpoints
Strengths
- Contains 4 distinct model checkpoints corresponding to a 4-fold training split
- Utilizes the vit_huge_plus_patch16_dinov3 architecture for high-resolution feature extraction
- Validated performance with a 0.74 silver medal score in a biomass prediction competition