search

AaronCUG / Deep-pyramid-similarity-model

1 stars 0 forks source link

Inquiry About best.pth File #1

Open Yingjie1228 opened 10 months ago

Yingjie1228 commented 10 months ago

Hello,

I am currently working with your project. I noticed that the script expects a pre-trained model file named best.pth, but I could not find this file in the repository.

Could you please provide some guidance on how to obtain this file? Here are a few specific questions that I have:

  1. Is the best.pth file available for download? If so, could you please provide the link or the instructions on how to access it?

  2. If the file is not available, could you provide some instructions or guidelines on how to train the model and generate this file? Any details about the training procedure, including the dataset used, would be extremely helpful.

  3. Are there any alternative pre-trained models or weights that can be used with this script?

I appreciate the work you've put into this project, and any assistance you can provide would be greatly appreciated.

Thank you for your time and support.

AaronCUG commented 10 months ago

Hi there,

Thanks for your inquiry. Unfortunately, the "best.pth" is not available for downloading. Due to the variation in identification of urban redevelopment, we are concerned that sharing the "best.pth" possibly has unexpected side effects on subsequent research. However, it is simple to train this model.

  1. Prepare training dataset. The training dataset should contain street view image (SVI) pairs of a property and whether it's redeveloped (e.g., 1 or 0). I used 2000 SVI pairs in my training. More samples potentially further improve the performance of the model.
  2. Write a code snippet of training process. During my training, the parameters of VGG16 were fixed and initialized by a pretrained VGG16 model from PyTorch. Only the parameters of fully connected layers were optimized. There is no alternative pretrained model for the fully connected layers.
  3. Save the "best.pth" file. Once your model achieves a satisfactory result, the model and its optimized parameters can be saved for ensuing inference.

More details can be found in the associated paper: https://doi.org/10.1016/j.cities.2023.104644 Let me know if you have any further questions.