Datasets:
| { | |
| "model": "ResNet-50 (RadImageNet)", | |
| "loss_function": "CE", | |
| "must_include": true, | |
| "learning_rate": 1e-05, | |
| "batch_size": 16, | |
| "weight_decay": 0.0001, | |
| "optimizer": "AdamW", | |
| "scheduler": "CosineAnnealing", | |
| "image_augmentations": { | |
| "rotation": 15, | |
| "horizontal_flip": true, | |
| "random_crop": true, | |
| "color_jitter": true, | |
| "brightness": 0.2, | |
| "contrast": 0.2, | |
| "imagenet_norm": true | |
| }, | |
| "dropout": 0.2, | |
| "loss_specific_params": "class_weights=inverse_frequency", | |
| "multi_target_strategy": "Shared backbone + task-specific heads", | |
| "single_target_strategy": null, | |
| "pretrained_weights": "RadImageNet", | |
| "fine_tuning_strategy": "Full fine-tuning", | |
| "expected_gpu_memory": "6-8GB", | |
| "architectural_family": "Medical CNN", | |
| "class_weighting": "inverse_frequency", | |
| "sampling_strategy": "balanced_batch", | |
| "threshold_selection": "F1_optimal", | |
| "experiment_name": "ResNet-50_RadImageNet_lr1e-05_bs16" | |
| } | |