If you want to test weakly-supervised object detection on your own data, we provide some tips to ease the process.
You can also configure your own paths to the datasets.
For that, all you need to do is to modify wetectron/config/paths_catalog.py to
point to the location where your dataset is stored.
You can also create a new paths_catalog.py file which implements the same two classes,
and pass it as a config argument PATHS_CATALOG during training.
You can use alternative methods such as Edge Boxes for computing proposals. To create a proposal file, you need to save a pickle file with two keys boxes and indexes.
Note that the indexes have to be consistent with your dataset.
Please check the provided proposal files for details. Use encoding="latin1 as loading flag:
with open('MCG-coco_2014_minival-boxes.pkl', 'rb') as f:
proposals = pickle.load(f, encoding="latin1")
In your configuration files, set MODEL.ROI_BOX_HEAD.NUM_CLASSES to the number of classes of your dataset.
Change DATASETS and PROPOSAL_FILES to the corresponding files generated from the above steps.
Tunning the hyperparameters in SOLVER.* based on your validation results.
Feel free to post the issues you meet during the process!