F2Wang
to get highter precision dataset
@F2Wang Thank you for your excellent work. I got unsatisfactory preformance when training singlehotpose with my own dataset. I guess it may be related to data annotation, I want you to give me some advice
- is there any skills for data annotation? Such as trim the mesh file
- except record.py and record2.py, have you tried to create datasets with higher precision parameters using realsense, such as with higher resolution?
I hope your problem has been resolved by now. I am not sure what you mean by having unsatisfactory performance when training singleshotpose with your own dataset, do you mean that you have a low train/validation accuracy? That shouldn't be the case if you initialized the training with some of their available pre-trained weights.
If you want to reduce overfit, you may want to try more aggressive data augmentation, and other skills used in reducing overfit in training.
I don't think using higher resolution will help. The original LINEMOD dataset is not of very high resolution, and the cnn methods usually resize the image anyway.
@F2Wang
First of all thanks for this wonderful setup for creating the dataset.
Can you let me know what tool I need to use to augment the data? Any help would be great. Thanks!
