dulucas/UVO_Challenge

How do you expect to use these Config and what is your replay process

WisleyWang opened this issue · 2 comments

WisleyWang commented

Can you explain your training process in detail? Because I see you have a lot of config files, I don't know which ones to use and how to use them

dulucas commented

Hi, as our code is based on MMDetection and MMSegmentation, the training code uses the same style. An example.

As for the configs, as specified in README:

  • For detection model pretraining: we use this
  • For detection model finetuning: we use this
  • For segmentation model pretraining: we use this
  • For segmentation model finetuning: we use this
  • For segmentation model inference: we use this
dulucas commented

The training process is:
For detection:

  • Pretrain on COCO
  • Fintune on UVO

For segmentation:

  • Pretrain on COCO, OpenImage and PASCAL(an ensemble of dataset)
  • Finetune on UVO