Skip to content

Latest commit

 

History

History
28 lines (22 loc) · 1.86 KB

File metadata and controls

28 lines (22 loc) · 1.86 KB

Model weights

Inside this folder models for the trained models are stored. Due to the size of the models, the models can be found on link:

  • resnet18-imagenet.pth - A pretrained model of the ResNet-18 architecture. Pretrained on ImageNet database,

  • resnet50-imagenet.pth - A pretrained model of the ResNet-50 architecture. Pretrained on ImageNet database,

  • resnet101-imagenet.pth - A pretrained model of the ResNet-101 architecture. Pretrained on ImageNet database,

  • Weights for the segmentation module trained for semantic segmentation of 150 different classes found in the ADE20K database (directory: Pretrained model on 150 classes):

    • encoder_epoch_20.pth - Weights for the encoder after training the model for 20 epochs,
    • decoder_epoch_20.pth - Weights for the decoder after training the model for 20 epochs.
  • Weights for the segmentation module after transfer learning for semantic segmentation of walls. Transfer learning is done only for the last layer of the decoder architecture (directory: Transfer learning - last layer):

    • Output_only_encoder.pth - Weights for the encoder,
    • Output_only_decoder.pth - Weights for the decoder.
  • Weights for the segmentation module after transfer learning for semantic segmentation of walls. Transfer learning is done for the entire decoder architecture (directory: Transfer learning - entire decoder):

    • transfer_encoder.pth - Weights of the encoder,
    • transfer_decoder.pth - Weights of the decoder.
  • Weights for the segmentation module trained for semantic segmentation of only walls (The training is done on a subset of the ADE20K database) (directory: Without transfer learning):

    • best_encoder_epoch_19.pth - Weights of the encoder,
    • best_decoder_epoch_19.pth - Weights of the decoder.