Open mvstu opened 3 days ago
Thank you for your attention. Regarding predicting your own data, you can refer to the relevant tutorials of MMDetection to organize your data in COCO format, modify the corresponding test data parameters in the configs file, and then run the test script. As for how to process the predicted .pkl
file, you can refer to the relevant code in tools/bonai/bonai_evaluation.py
.
Thank you for your patient answer! Also, may I ask a question? Is it okay if my test image is not labeled? So, using the training model from Beijing and Xi'an to predict my data? At the same time, when I was running your test script, I was missing the file 'bonai_sthanghai_xian_test_roof.json'. I checked the data and the original file 'bonai/bonai_ievalation-py', but I only have the file 'bonai_sthanghai_xian_test.json'. May I ask why this is happening? How can I obtain this file? Thank you for your answer!
.json
file containing only filenames without semantic annotations. This .json
file is used to specify the files for testing. Then, you can run the test script without the --eval
parameter, so it only performs prediction without evaluation, i.e., inference.bonai_shanghai_xian_test_roof.json
means that the annotations such as 'segmentation' and 'bbox' in the file correspond to the roof, as the code by default reads annotation information from these two fields. Similarly, bonai_shanghai_xian_test_footprint.json
contains annotations for the building's footprint. You can convert it based on the fields 'footprint_mask', 'footprint_bbox', 'roof_bbox', and 'roof_mask' in bonai_shanghai_xian_test.json
.Thank you for your detailed and patient answer! Let me try it, thank you! Wishing you a happy life and smooth scientific research!
Hello, thank you for your excellent work and sharing! May I ask if I am able to predict my own data and how to handle the pkl file obtained? Thank you!