| image_root: '' # for image path | |
| ann_root: '' # for json path | |
| dataset: 'facecaption' | |
| config: './configs' | |
| pretrained: '' | |
| intermediate_hidden_state: False | |
| # size of vit model; base or large | |
| vit: 'base' | |
| image_size: 224 | |
| batch_size_train: 80 | |
| batch_size_test: 80 | |
| queue_size: 61440 | |
| alpha: 0.4 | |
| k_test: 256 | |
| # optimizer | |
| weight_decay: 0.05 | |
| init_lr: 3e-5 | |
| min_lr: 1e-6 | |
| warmup_lr: 1e-6 | |
| lr_decay_rate: 0.9 | |
| max_epoch: 15 | |
| warmup_steps: 20000 | |