This GitHub Repository was produced to share material relevant to the Journal paper "Automatic crack classification and segmentation on masonry surfaces using convolutional neural networks and transfer learning" by D. Dais, İ. E. Bal, E. Smyrou, and V. Sarhosis published in "Automation in Construction".
I'm currently working on reproducing the results of your research. Using the complete original dataset, I am only able to get a dilated F1-score of 73.5% using the Unet model and Mobilenet backbone. All training parameters have been set using the parameters from the paper, so I'm wondering what could cause the difference.
To make my question concrete: Did you apply any kind of extra steps not mentioned in the paper in order to achieve the high F1-score mentioned? Or alternatively: are there major differences in this repo which can cause a trained model using this code to perform worse? I understand that you might not have the time to look into this, so I'm also fine with you just sending the original code if you still have it so I can solve this problem myself. Thank you in advance for your time!
Hi @dimitrisdais ,
I'm currently working on reproducing the results of your research. Using the complete original dataset, I am only able to get a dilated F1-score of 73.5% using the Unet model and Mobilenet backbone. All training parameters have been set using the parameters from the paper, so I'm wondering what could cause the difference.
To make my question concrete: Did you apply any kind of extra steps not mentioned in the paper in order to achieve the high F1-score mentioned? Or alternatively: are there major differences in this repo which can cause a trained model using this code to perform worse? I understand that you might not have the time to look into this, so I'm also fine with you just sending the original code if you still have it so I can solve this problem myself. Thank you in advance for your time!
Best regards,
David