foolwood / DCFNet_pytorch

DCFNet: Discriminant Correlation Filters Network for Visual Tracking
https://arxiv.org/pdf/1704.04057.pdf
MIT License
208 stars 60 forks source link

the DCFnet seems can't converge #20

Open xxAna opened 4 years ago

xxAna commented 4 years ago

Thank you for your excellent work and your share! These days,I copy the code and retrain the DCFnet, to my surprise,I find that although the mean loss is decreasing, the loss seems fluctuating.Just like below. As I didn't change key parts of the DCFnet. It confused me.Do you ever met a similar phenomenon? And if it convenient to give me some advices? Sorry for my poor English and thank you again. image

poppinjie commented 4 years ago

I meet the same problem,have you solved the problem?If so,please help me work it out.