Details about the traning
See original GitHub issueNice job. I am trying to reproduce your work with mmdetection
and before it could you help me to confirm some details?
- The input scale of the training images. Does line https://github.com/facebookresearch/unbiased-teacher/blob/6977c6f77c812fae4064dc1b3865658c2ed247b1/configs/Base-RCNN-FPN.yaml#L41 indicate the size of the input image is selected from those scales randomly?
- The total batch size for training. The batch size is 32 labeled images + 32 unlabeled images. And For each iteration, 96 images( 32 strong augmented labeled images + 32 weakly augmented labeled images + 32 strong augmented unlabeled images) are used for supervision, right?
Issue Analytics
- State:
- Created 2 years ago
- Comments:6 (3 by maintainers)
Top Results From Across the Web
13 tips: Training descriptions that give you more participants
How to get more participants? Step 1: Write good training descriptions. Here are 13 tips for writing a good training description!
Read more >Training - Wikipedia
Training is teaching, or developing in oneself or others, any skills and knowledge or fitness that relate to specific useful competencies.
Read more >What is training and development? | IBM
Training and development programs help employees learn and acquire new skills and gain the professional knowledge required to progress their careers.
Read more >Meaning, Definition and Types of Training - Your Article Library
Training refers to the teaching and learning activities carried on for the primary purpose of helping members of an organization acquire and apply...
Read more >10 Examples of Training Programs and Their Benefits - Indeed
Employees who receive extensive training that prepares them for the job often report higher job satisfaction overall. Learning about different ...
Read more >Top Related Medium Post
No results found
Top Related StackOverflow Question
No results found
Troubleshoot Live Code
Lightrun enables developers to add logs, metrics and snapshots to live code - no restarts or redeploys required.
Start FreeTop Related Reddit Thread
No results found
Top Related Hackernoon Post
No results found
Top Related Tweet
No results found
Top Related Dev.to Post
No results found
Top Related Hashnode Post
No results found
Top GitHub Comments
Yes. We have tried to apply the weight to all samples but the improvement is marginal compared to only apply to the background part.
Got it. Did you try to apply the teacher’s predicted weight to foreground samples before?