The results of the test model did not meet expectations？ about siamdw HOT 11 CLOSED

Thanks for your code fisrt, it help so much.
According to the configuration process you said, I successfully ran the test code.
I then converted the txt files into a mat files and put it into the OTB framework to calculate the result. However, I found that the SiamFCRes22 model only reached 0.615 on the OTB2013. So I am very confused about what is the reason.

1. check your python packages environment again
2. check your translation code
3. the reader in closed issuse #2 has reproduced the result. May there is mistake with some segments during your reproducing.

from siamdw.

Thanks for your code fisrt, it help so much.
According to the configuration process you said, I successfully ran the test code.
I then converted the txt files into a mat files and put it into the OTB framework to calculate the result. However, I found that the SiamFCRes22 model only reached 0.615 on the OTB2013. So I am very confused about what is the reason.

Hello~
Have you fixed the problem? If yes, this issue will be closed.

from siamdw.

@haizhongli Have you fixed the problem？We had test the reproducibility of our training and test code. Also, Lots of other githubers said that they reproduced the results. If you have any problems, pls ping us.

from siamdw.

from siamdw.

from siamdw.

@haizhongli Have you reproduce the results by our code?

from siamdw.

from siamdw.

from siamdw.

Dear researcher: When I looked at the code again, I found that the only trainable parameter for training was the second residual block, features. features. layer2. Did you set it up on purpose? If not, maybe in my environment, there are problems in this place.
…
------------------ 原始邮件 ------------------ 发件人: "Houwen Peng"[email protected]; 发送时间: 2019年6月23日(星期天) 中午1:05 收件人: "researchmm/SiamDW"[email protected]; 抄送: "﹎╁划濄∑淚"[email protected];"Mention"[email protected]; 主题: Re: [researchmm/SiamDW] The results of the test model did not meet expectations？ (#12) @haizhongli Have you reproduce the results by our code? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

1. layer2 means the third stage of resnet.
2. Which training policy do you use now? Unfix all layer at once or gradually? (You can find details in backbone.py)
3. Trainable parameter is only layer2 through all 50 epoch or only at the begining of trainig?
4. What's the performance after tuning?

from siamdw.

I haven't noticed the fine-tuning steps before, this will have an impact, the author has been patiently following up on my problem, then I will try para-tune. Any questions, please ask again, thank you

from siamdw.

According to the instructions in readme, the parameters provided by the same environment test author are configured, and the results are consistent with the publication. Thank you for your patient answer.

from siamdw.