andreluizbvs / plad Goto Github PK

Thanks for providing a piece of detailed power line assets dataset. However, I can't reach appropriate average precision when i use PLAD dataset. I find that you describe that you split in a standard 80/20 proportion for the training and test sets in the paper. And the dataset folder in github include two sub-folders named 20181127-A1 and 20181129-R. So i want to know whether 20181127-A1 is train sub-folder and 20181129-R is test subfolder.

If not, Can you send me a piece of exact dataset that have splited train and test sets? The following pictures introduce these problems.

Hello, After reading your article, I think the project is very helpful for asset intelligent detection, and I really want to reproduce your code.Thank you very much!You can email me at [email protected]

Please, is your dataset in coco format？

Please how can I get the code # # @andreluizbvs

When I started parsing the dataset, I saw this in annotations.json (annotations['categories']):

[{'supercategory': 'tower', 'id': 0, 'name': 'tower'}, {'supercategory': 'component', 'id': 1, 'name': 'insulator'}, {'supercategory': 'component', 'id': 2, 'name': 'spacer'}, {'supercategory': 'component', 'id': 3, 'name': 'damper'}, {'supercategory': 'component', 'id': 5, 'name': 'plate'}]

The IDs start at 0 and finish at 5 that's 6 classes, or we have 5 classes. This also extends to the annotations['images'].

I think it should be: 0,1,2,3,4 or probably I'm missing something...

Can you give me more clarification? Thanks.

BTW I downloaded the dataset from Google Drive.