search

xuebinqin / U-2-Net

The code for our newly accepted paper in Pattern Recognition 2020: "U^2-Net: Going Deeper with Nested U-Structure for Salient Object Detection."
Apache License 2.0
8.44k stars 1.45k forks source link

Unexpected item included in the final mask #50

Open ohheysherry66 opened 4 years ago

ohheysherry66 commented 4 years ago

Congratulations for your amazing work with u2net！ Recently,I have tried your net on portrait segmentation task.Also I trained it from scratch by portrait segmentation dataset without other objects; It got great performance.However,I find that sometimes,objects such as chairs and street nameplates can also be included. I am confused about it. Since I trained it from scratch by the given dataset.It can be seen as a semantic segmentation task,right?Why the other object can be inclued? Thankyou!

PytaichukBohdan commented 4 years ago

Congratulations for your amazing work with u2net！ Recently,I have tried your net on portrait segmentation task.Also I trained it from scratch by portrait segmentation dataset without other objects; It got great performance.However,I find that sometimes,objects such as chairs and street nameplates can also be included. I am confused about it. Since I trained it from scratch by the given dataset.It can be seen as a semantic segmentation task,right?Why the other object can be inclued? Thankyou!

Could you please share the process of your training? Did you freeze layers (which ones)? Which parameters & optimizer did you choose? Maybe some other tricks?

Thanks in advance!

ohheysherry66 commented 4 years ago

Congratulations for your amazing work with u2net！ Recently,I have tried your net on portrait segmentation task.Also I trained it from scratch by portrait segmentation dataset without other objects; It got great performance.However,I find that sometimes,objects such as chairs and street nameplates can also be included. I am confused about it. Since I trained it from scratch by the given dataset.It can be seen as a semantic segmentation task,right?Why the other object can be inclued? Thankyou!

Could you please share the process of your training? Did you freeze layers (which ones)? Which parameters & optimizer did you choose? Maybe some other tricks?

Thanks in advance!

Hi,I just use the original optimizer the writer provided in the code.I use the pretrained model and add the other layer to predict boundary by freeze all parameters and random init the added layer,but it seems help a little.