search

nka77 / DAHiTra

DAHiTra: UNET Architecture with Hierarchical Transformers for Automated Building Damage Assessment Using Satellite Imagery
MIT License
26 stars 6 forks source link

weights #6

Open Tamer-Saleh opened 1 year ago

Tamer-Saleh commented 1 year ago

Nice work and congratulations on the publication.

Could you share the model weights of BASE_UNet_Transformer_img1024_lossOrig_alldata to train the mode with my own dataset? Here is my email for further cooperation, tamersaleh@whu.edu.cn.

Thank you!

mmuneeburahman commented 1 year ago

Trained weights for the xBD dataset from here. The training was done for 120 epochs. You can find the logs here

shah0nawaz commented 11 months ago

Trained weights for the xBD dataset from here. The training was done for 120 epochs. You can find the logs here

Thanks for sharing the weights

m3zai6i commented 11 months ago

Trained weights for the xBD dataset from here. The training was done for 120 epochs. You can find the logs here

Thanks for sharing the trained weights you saved our training time.

shah0nawaz commented 10 months ago

@mmuneeburahman hi, I tried your pretrained weights. and test it i am unable to reproduce the results. The results are very noisy and has no meaning. I am sure i am doing something wrong. Please if you can give me a direction so that I can start from there. I have also trained "xBD_base_transformer_pos_s4_dd8_o5" model. but unable to reproduce the results. My training process is good and the numbers which the training reproduces are fair enough but testing is giving me total noise

These are my training and test log fils log.txt log_test.txt

shah0nawaz commented 10 months ago

@mmuneeburahman hi, I tried your pretrained weights. and test it i am unable to reproduce the results. The results are very noisy and has no meaning. I am sure i am doing something wrong. Please if you can give me a direction so that I can start from there. I have also trained "xBD_base_transformer_pos_s4_dd8_o5" model. but unable to reproduce the results. My training process is good and the numbers which the training reproduces are fair enough but testing is giving me total noise

These are my training and test log fils log.txt log_test.txt

Please give me your guidance on this issue.