Model-J: DINO Model (model_idx_0510)
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 |
| Epochs | 6 |
| Max Train Steps | 1998 |
| Batch Size | 64 |
| Weight Decay | 0.05 |
| Seed | 510 |
| Random Crop | True |
| Random Flip | False |
Performance
| Metric | Value |
|---|
| Train Accuracy | 0.3312 |
| Val Accuracy | 0.2987 |
| Test Accuracy | 0.3020 |
Training Categories
The model was fine-tuned on the following 50 CIFAR100 classes:
couch, streetcar, crab, tank, man, pear, butterfly, crocodile, skyscraper, boy, palm_tree, chair, cup, girl, telephone, table, cockroach, camel, lawn_mower, rocket, tractor, pickup_truck, whale, lizard, turtle, cloud, trout, mountain, baby, sunflower, shark, snail, beaver, poppy, sea, hamster, worm, maple_tree, wolf, apple, can, ray, snake, mouse, bus, elephant, raccoon, motorcycle, castle, leopard
