Model-J: DINO Model (model_idx_0362)
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 | constant_with_warmup |
| Epochs | 5 |
| Max Train Steps | 1665 |
| Batch Size | 64 |
| Weight Decay | 0.009 |
| Seed | 362 |
| Random Crop | True |
| Random Flip | True |
Performance
| Metric | Value |
|---|
| Train Accuracy | 0.3298 |
| Val Accuracy | 0.3120 |
| Test Accuracy | 0.3140 |
Training Categories
The model was fine-tuned on the following 50 CIFAR100 classes:
house, bottle, cloud, camel, aquarium_fish, boy, forest, lobster, plate, turtle, spider, seal, tractor, wardrobe, dolphin, flatfish, couch, porcupine, snail, woman, tulip, leopard, telephone, fox, worm, bowl, skyscraper, ray, bicycle, whale, lion, lawn_mower, hamster, tiger, rose, train, bridge, snake, castle, orange, bee, sea, sunflower, mushroom, lamp, can, crab, maple_tree, apple, bear
