RodrigoGantier/Mask_R_CNN_Keypoints

How to train my own dataset?

zhoulukuan opened this issue · 5 comments

Thanks for your great work.
I want to train my own dataset, which has 5 classes and more keypoints. Can you give me some advice on how to change your code?
I have read main.py file and change ai_data_set() to load my data. However, I don't konw the meaning and use of variable class_mask in load_mask().
Besides, do I need to change code in other files such as model.py and utils.py due to the different numbers of classes and keypoints?

@zhoulukuan You can refer to my repository https://github.com/Superlee506/Mask_RCNN. I refereed to the original detectron project and modified the code with detailed comments. The loss converges quickly, but there is still much room for improvement.

@Superlee506 Thanks a lot. In fact, I want to use mask rcnn to predict keypoints in clothes. I have two questions:

  1. Due to the different formats of train/test data, I must change it to the format of COCO annotations, right?
  2. I have more classes and keypoints. Do I need to change the code of network/loss or other places? Can you give me some help?
    Thank you in advance.

@zhoulukuan You can post the question under my repository and I will answer it as soon as I can.

@Superlee506 Sorry, I cannot find the ‘issue’ in your repository. Do you open the issue function in your setting?

@zhoulukuan Sorry, It's ok now.