Skip to content

DSC结果 #72

@xifeige

Description

@xifeige

Hello author, I trained ISIC2018 myself on a 4090 with CUDA 11.8 using the same 250 epochs.

  • Test of the best model, loss: 0.3125, miou: 0.7811195049475715, f1_or_dsc: 0.8771107191603792, accuracy: 0.9497523674598107, specificity: 0.9708718337536931, sensitivity: 0.868574568513323, confusion_matrix: [[26255387, 787716], [924659, 6110958]]
  • Using the official weights you provided, UltraLight_VM_UNet.pth
  • Test of the best model, loss: 0.2997, miou: 0.786355696204604, f1_or_dsc: 0.8804021482119618, accuracy: 0.9540482741135817, specificity: 0.9891206641486371, sensitivity: 0.819239023386293, confusion_matrix: [[26748892, 294211], [1271765, 5763852]]

It still can't reach the 0.894 reported in the paper. Could you give me some advice, or is this just normal fluctuation?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions