Model-J: DINO Model (model_idx_0327)
This model is part of the Model-J dataset, introduced in:
Learning on Model Weights using Tree Experts (CVPR 2025) by Eliahu Horwitz*, Bar Cavia*, Jonathan Kahana*, Yedid Hoshen
ProbeXModel Details
| Attribute | Value |
|---|
| Subset | DINO |
| Split | train |
| Base Model | facebook/dino-vitb16 |
| Dataset | CIFAR100 (50 classes) |
Training Hyperparameters
| Parameter | Value |
|---|
| Learning Rate | 0.0005 |
| LR Scheduler | cosine_with_restarts |
| Epochs | 6 |
| Max Train Steps | 1998 |
| Batch Size | 64 |
| Weight Decay | 0.03 |
| Seed | 327 |
| Random Crop | False |
| Random Flip | True |
Performance
| Metric | Value |
|---|
| Train Accuracy | 0.4681 |
| Val Accuracy | 0.3861 |
| Test Accuracy | 0.3922 |
Training Categories
The model was fine-tuned on the following 50 CIFAR100 classes:
lamp, tiger, wardrobe, wolf, bowl, cloud, road, sunflower, keyboard, bee, crab, streetcar, mushroom, squirrel, pear, whale, couch, cockroach, worm, kangaroo, poppy, man, bus, sweet_pepper, dinosaur, clock, bottle, oak_tree, plate, lawn_mower, butterfly, shrew, skunk, bridge, palm_tree, mountain, baby, castle, snake, spider, ray, sea, porcupine, trout, aquarium_fish, caterpillar, seal, bicycle, house, beetle
