README.md

Variables and Parameters in the Configuration File (in MIAOD.py)

Language: 简体中文 | English

• Base
• Datasets
• Model Training
• Results
• Unique MI-AOD Settings

Base

• _base_ (Lines 4-7): The basic configuration files, including RetinaNet, PASCAL VOC and runtime settings.

Datasets

• data_root (Line 2): Your actual directory. There is another same data_root in _base_/voc0712.py to modify.

• data (Lines 9-16): The data used for training, validation and test.

  • test (Line 10): The data for test.

    • ann_file (Lines 11-14): The path of annotation file.

    • img_prefix (Line 15): The path of images.

Model Training

• model (Line 17): The used model, including backbone, neck and head.

  • bbox_head: The bounding-box head of the model.

    • C: The number of classes in the dataset.

• optimizer (Line 19): The used optimizer, including learning rate, momentum and weight decay.

  • lr: Learning rate.

  • momentum: Momentum parameter.

  • weight_decay: Weight decay parameter.

• optimizer_config (Line 20): The gradient harmonizing parameter.

• lr_config (Line 22): The settings in learning rate.

  • step: The moment to drop the learning rate.

Results

• checkpoint_config (Line 24): The frequency of saving models.

• log_config (Line 26): The frequency of printing training logs.

• evaluation (Line 31): The frequency of evaluating the model.

• work_directory (Line 48): The work directory for saving logs and files. Please refer to here for more information.

Unique MI-AOD Settings

• epoch_ratio (Line 29): The number of epochs for Label Set Training step and those for Re-weighting and Minimizing/Maximizing Instance Uncertainty steps.

• epoch (Line 33): The number of outer loops (i.e., all 3 training steps except the first Label Set Training step).

• X_L_repeat (Line 36): The repeat time for the labeled sets.

• X_U_repeat (Line 37): The repeat time for the unlabeled sets.

• train_cfg (Line 39): Some parameters for model training.

  • param_lambda: The regularization hyper-parameter .

• k (Line 40): The hyper-parameter k, the number of observed top instance uncertainty in an unlabeled image for sample selection.

• X_S_size (Line 43): The size of , the newly selected sets.

• X_L_0_size (Line 44): The size of , the initial labeled set.

• cycles (Line 46): The number of active learning cycles.