Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

PPC pretrained weights #20

Open
trungpham2606 opened this issue Feb 26, 2020 · 12 comments
Open

PPC pretrained weights #20

trungpham2606 opened this issue Feb 26, 2020 · 12 comments

Comments

@trungpham2606
Copy link

❓ Questions and Help

@mrlooi
Can you public your pretrained weights on PPC dataset ?

@mrlooi
Copy link
Owner

mrlooi commented Feb 26, 2020

Hey unfortunately I don't have the pretrained weights with me. In any case the MSCOCO pretrained weights should be more better - it's more general

@trungpham2606
Copy link
Author

@mrlooi
In addition,
You have not published your paper yet, have you ?

@mrlooi
Copy link
Owner

mrlooi commented Feb 26, 2020

Nope, it takes work, and doubt it's going to make much of an impact anyway. I'm not a full time researcher, so publications and citations are not as important to me xD

@trungpham2606
Copy link
Author

@mrlooi
I often meet this error after some training iterations:
image
Can you tell me how to solve this issue ?

@mrlooi
Copy link
Owner

mrlooi commented Feb 26, 2020

It would seem that the input batch size is empty (0 x 1 x ...). A good way to check is backtrack where the input became empty. I haven't seen this issue before, but if you do find out the cause, please let me know

@trungpham2606
Copy link
Author

@mrlooi
Yeah sure, Iam training with my own dataset to see if this error still occurs.

@trungpham2606
Copy link
Author

@mrlooi
I think the error comes from the model not about the dataset.
For PPC dataset, i had printed out the images when the errors come out.
image
image

The error comes with different images each times.
The objects inside the images I guess they're not too small.
Do you have any ideas about this ?

@mrlooi
Copy link
Owner

mrlooi commented Feb 26, 2020

Hey good question. I think it could be that the ground truth rotated bboxes for those objects did not have good IoU fit with any of the rotated anchors

@trungpham2606
Copy link
Author

@mrlooi
I had changed some threshold for being able to train PPC dataset.
Thank you for your prompt support!

@trungpham2606
Copy link
Author

trungpham2606 commented Feb 28, 2020

@mrlooi
Excuse me but can you please give me some sources of reference you had used in your work ?
Does RRPN you used come from this paper: Arbitrary-Oriented Scene Text Detection via
Rotation Proposals ?(https://arxiv.org/pdf/1703.01086.pdf)

@mrlooi
Copy link
Owner

mrlooi commented Feb 28, 2020

Yes RRPN and MaskRCNN are the main ones

@elepherai
Copy link

@mrlooi
I had changed some threshold for being able to train PPC dataset.
Thank you for your prompt support!

@trungpham2606 Hi, could you tell me what thresholds did you change for training PPC dataset? Thanks~

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants