could you please tell me about the partition proportion of training, validation, and testing dataset.

[chriszxk]

Hi:

could you please tell me about the partition proportion of training, validation, and testing dataset.

Thanks,
chris

[VincentHancoder]

Hello, the proportion we use to divide dataset is in the ratio of 4:1:1, we use the validation dataset as testing dataset as well. Meanwhile, you can check the same result of train files and corresponding validation files in annotation folder(e.g. train1.json and val1.json).

[chriszxk]

Hello, the proportion we use to divide dataset is in the ratio of 4:1:1, we use the validation dataset as testing dataset as well. Meanwhile, you can check the same result of train files and corresponding validation files in annotation folder(e.g. train1.json and val1.json).

So, is the accuracy (AP.5, AP75) computed on validation dataset in your paper,

[VincentHancoder]

Yes, and we use the 5-fold cross-validation method to obtain and calculate AP values.

[chriszxk]

Yes, and we use the 5-fold cross-validation method to obtain and calculate AP values.

Hello, I noticed that there is a class called "OK" in the dataset. I would like to know if you included the "OK" objects in the training process.

[hs19991218]

请问在训练当中是如何解决7种缺陷数量不均衡的问题呢？