search

IlchaeJung / RT-MDNet

103 stars 30 forks source link

The results of OTB2015 and training time #10

Open yangkai12 opened 5 years ago

yangkai12 commented 5 years ago

Thank you for releasing this excellent work. I have two questions to ask you.

  1. I tested your model on OTB2015 and the result was 0.632,This result is lower than 0.650 in your paper. I use the default parameter settings of your code. Could you please tell me how to solve this issue?
  2. I train ImageNet-Vid dataset on a GeForce RTX 2080, how long will it take? Thank you very much.
shipubupt commented 5 years ago

I also meet the same problem 1...

TOT-BENCHMARK commented 5 years ago

me too

IlchaeJung commented 5 years ago

Thank you for your interests.

For first problem, This is the characteristics of stochastic algorithms. In Deep learning based approach like MDNet, TCNN, RT-MDNet, etc, we stochastically draw samples many times for estimate, update. Therefore, for every trials, the results may be different from each other. However, in our algorithm the variance of success rate is within 0.02~0.03. Therefore, if you obtain 0.632 as a score of SR, you can obtain the score from paper in additional trial. Thank you :)

Secondly, I am not sure, but, when I use Titan X, it takes more than 1 week. Maybe, with RTX 2080, it will be longer than my case.

yangkai12 commented 5 years ago

Okay, thank you for your reply. I understand these two questions.

Tomingz commented 5 years ago

I use python2.7 and pytorch0.2 or pyto0.4, GPU 940MX, ONLY 2GB,there is always this error:cudaCheckError() failed : invalid device function. How to solve it?thank u.