Hang Chen

Please refer to https://github.com/tinyalpha/BPR#inference

This shows that the patch with id 16 is not included in `refined.pkl`. You may check if the `refined.pkl` is consistent with `patches/detail_dir/val`.

You can refer to #15 for visualization

The difference between them is that the former only process the validation set and the latter also process the training set. For the validation set, both do the same thing.

The IOU_THRESHs of training and validation sets don't need to be the same. But the same validation set should the same throughout the inference. E.g. if you generated the `refined.pkl`...

HI, we generated those json files using mmdetection. You can refer to the 6th example in https://mmdetection.readthedocs.io/en/stable/1_exist_data_model.html#examples.

Hi, for question 1, you can do this by slightly modifying the configure file of mmdet, for example: ```python data = dict( imgs_per_gpu=1, workers_per_gpu=2, test=dict( type=dataset_type, ann_file=data_root + # 'annotations/instancesonly_filtered_gtFine_test.json',...

Please find checkpoints at the following links: - trained on PointRend: https://cloud.tsinghua.edu.cn/f/a2497064f25949d693ba/?dl=1 - trained on SegFix: https://cloud.tsinghua.edu.cn/f/1befec256b8e4a02b2da/?dl=1

For the other question, we found that models trained on worse results typically had stronger refinement ability, and we guessed that the reason might be similar to data augmentation. We...

Hi, the config path and the number of GPU should be specified as arguments when calling dist_train.sh. You can refer to https://github.com/chenhang98/BPR#train-the-network for details.