Hey @ybkscht , I tried a lot of things but the ADD is not improving. I don't even see statements like "ADD improved/ ADD did not improve". The model does get better though.
The ADD value mostly remains 0 or changes to 0.0010 or some other small value.
Can you suggest some direction?
I have also tried for a large dataset like 10,000 synthetically generated images but I don't see a nice performance in the real world. What do you think should be the number of epochs to get a good real-world performance, when starting with pre-trained Linemod or Occlusion dataset weights?
Hey @ybkscht , I tried a lot of things but the ADD is not improving. I don't even see statements like "ADD improved/ ADD did not improve". The model does get better though.
The ADD value mostly remains 0 or changes to 0.0010 or some other small value. Can you suggest some direction?![image](https://user-images.githubusercontent.com/39311993/163452522-bf135782-1e96-4ffb-8a58-665d8befc15e.png)
I have also tried for a large dataset like 10,000 synthetically generated images but I don't see a nice performance in the real world. What do you think should be the number of epochs to get a good real-world performance, when starting with pre-trained Linemod or Occlusion dataset weights?
Thanks!