pointSIFT
pointSIFT copied to clipboard
MVIG-SJTU
Can anyone get the results of scannet in paper?
I have not changed the code of pointsift and I can't get the result reported in paper,The result in paper is 86.1, I can only get 84.8, Can someone get the results of scannet in paper?Maybe I made some mistake,Thank you!
Yes, and after some modification I can get 86.3. You can check in your tensorboard.
@sjtuytc can you describe your modification? And, how many GPU you use to train the model to get the 86.3? Thanks a lot!
@sjtuytc Could you please describe your modification? I both use 2 or 4 GPU but I can only get 85.42.
I add more shortcuts, and the modified network model is with the attachment.
----- 原始邮件 ----- 发件人: "WuzhaoLi" [email protected] 收件人: "MVIG-SJTU/pointSIFT" [email protected] 抄送: "sjtuytc" [email protected], "Mention" [email protected] 发送时间: 星期五, 2019年 1 月 11日 下午 3:12:55 主题: Re: [MVIG-SJTU/pointSIFT] Can anyone get the results of scannet in paper? (#19)
@sjtuytc Could you please describe your modification? I both use 2 or 4 GPU but I can only get 85.42.
-- You are receiving this because you were mentioned. Reply to this email directly or view it on GitHub: https://github.com/MVIG-SJTU/pointSIFT/issues/19#issuecomment-453403527
@sjtuytc oh~ Thanks for your reply. But I can't find the attachment everywhere 😂... Could you please send the modified network model again? Thanks a lot~
@sjtuytc oh~ Thanks for your reply. But I can't find the attachment everywhere 😂... Could you please send the modified network model again? Thanks a lot~
@sjtuytc hi, Thanks for your sharing the code. but I found the loss stopped decreasing at about 11. Could you tell me how you set the parameters? such as how many epoches you set?
One year has past and I could not remember those details, sorry. Only learning rate matters, I remember.
I have not changed the code of pointsift and I can't get the result reported in paper,The result in paper is 86.1, I can only get 84.8, Can someone get the results of scannet in paper?Maybe I made some mistake,Thank you!
Hi，I didn't change any code and I got calibrated vox accuracy around 84.6%, for vox accuracy I can only get 83.0%. Do you also get calibrated vox accuracy 84.8% with the original code?
I have not changed the code of pointsift and I can't get the result reported in paper,The result in paper is 86.1, I can only get 84.8, Can someone get the results of scannet in paper?Maybe I made some mistake,Thank you!
Hi，I didn't change any code and I got calibrated vox accuracy around 84.6%, for vox accuracy I can only get 83.0%. Do you also get calibrated vox accuracy 84.8% with the original code?
Note the reported result is the best acc over all epoches, not the last epoch.
I have not changed the code of pointsift and I can't get the result reported in paper,The result in paper is 86.1, I can only get 84.8, Can someone get the results of scannet in paper?Maybe I made some mistake,Thank you!
Hi，I didn't change any code and I got calibrated vox accuracy around 84.6%, for vox accuracy I can only get 83.0%. Do you also get calibrated vox accuracy 84.8% with the original code?
Note the reported result is the best acc over all epoches, not the last epoch.
Yes, I've noticed that, but the best vox average accuracy is still 83.0%. The best calibrated vox accuracy can reach to 84.6%. I didn't change any code and run on 4 gpus with 1000 epochs.
I have not changed the code of pointsift and I can't get the result reported in paper,The result in paper is 86.1, I can only get 84.8, Can someone get the results of scannet in paper?Maybe I made some mistake,Thank you!
Hi，I didn't change any code and I got calibrated vox accuracy around 84.6%, for vox accuracy I can only get 83.0%. Do you also get calibrated vox accuracy 84.8% with the original code?
Note the reported result is the best acc over all epoches, not the last epoch.
Yes, I've noticed that, but the best vox average accuracy is still 83.0%. The best calibrated vox accuracy can reach to 84.6%. I didn't change any code and run on 4 gpus with 1000 epochs.
Yes, you're right. You need to use my modified code.