search

ristea / sspcab

Other
147 stars 15 forks source link

Question about CutPaste #5

Open ghost opened 2 years ago

ghost commented 2 years ago

Thank you for your great work, I have a question: In the paper, you said, "For the methods [18, 34, 37, 39, 49, 79] chosen to serve as underlying models for SSPCAB, we use the official code from the repositories provided by the corresponding authors". However, it looks like the authors of CutPaste didn't release their official code. About the results from CutPaste+SSPCAB, I wonder if you reimplemented their code to generate the results?

ristea commented 2 years ago

Indeed the authors have not released an official implementation. Therefore, we used a public implementation from https://github.com/Runinho/pytorch-cutpaste

jianzhang96 commented 2 years ago

Thank you for your great work! In the public implementation CutPaste, the reproduced results (91.0 for classification) are lower than the paper (95.2). How did you fill the gap? In other words, your proposed SSPCAB can improve detection results from 91.0 to 96.1 with 5.1 performance gain.

BadSeedX commented 2 years ago

Thank you for your great work! In the public implementation CutPaste, the reproduced results (91.0 for classification) are lower than the paper (95.2). How did you fill the gap? In other words, your proposed SSPCAB can improve detection results from 91.0 to 96.1 with 5.1 performance gain. I have the same doubt

fighting-mumu commented 2 years ago

Thank you for your great work! In the public implementation CutPaste, the reproduced results (91.0 for classification) are lower than the paper (95.2). How did you fill the gap? In other words, your proposed SSPCAB can improve detection results from 91.0 to 96.1 with 5.1 performance gain. I have the same doubt

caiyu6666 commented 2 years ago

Thank you for your great work! In the public implementation CutPaste, the reproduced results (91.0 for classification) are lower than the paper (95.2). How did you fill the gap? In other words, your proposed SSPCAB can improve detection results from 91.0 to 96.1 with 5.1 performance gain. I have the same doubt

unrealgeometry commented 2 years ago

Thank you for your great work! In the public implementation CutPaste, the reproduced results (91.0 for classification) are lower than the paper (95.2). How did you fill the gap? In other words, your proposed SSPCAB can improve detection results from 91.0 to 96.1 with 5.1 performance gain. I have the same doubt

mjack3 commented 1 year ago

Thank you for your great work! In the public implementation CutPaste, the reproduced results (91.0 for classification) are lower than the paper (95.2). How did you fill the gap? In other words, your proposed SSPCAB can improve detection results from 91.0 to 96.1 with 5.1 performance gain. I have the same doubt