Hasano20's picture
Model save
745f767 verified
metadata
license: other
base_model: nvidia/mit-b5
tags:
  - generated_from_trainer
model-index:
  - name: SegFormer_mit-b5_Clean-Set3-Grayscale_Augmented_Hard
    results: []

SegFormer_mit-b5_Clean-Set3-Grayscale_Augmented_Hard

This model is a fine-tuned version of nvidia/mit-b5 on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 0.0143
  • Mean Iou: 0.9789
  • Mean Accuracy: 0.9908
  • Overall Accuracy: 0.9945
  • Accuracy Background: 0.9964
  • Accuracy Melt: 0.9810
  • Accuracy Substrate: 0.9951
  • Iou Background: 0.9930
  • Iou Melt: 0.9518
  • Iou Substrate: 0.9919

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 0.0002
  • train_batch_size: 8
  • eval_batch_size: 8
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_steps: 100
  • num_epochs: 25

Training results

Training Loss Epoch Step Validation Loss Mean Iou Mean Accuracy Overall Accuracy Accuracy Background Accuracy Melt Accuracy Substrate Iou Background Iou Melt Iou Substrate
0.3678 0.3030 50 0.1206 0.8584 0.9180 0.9591 0.9811 0.8082 0.9648 0.9560 0.6832 0.9361
0.1315 0.6061 100 0.0573 0.9293 0.9609 0.9808 0.9953 0.9068 0.9805 0.9764 0.8404 0.9710
0.0983 0.9091 150 0.0426 0.9427 0.9712 0.9855 0.9927 0.9330 0.9879 0.9865 0.8645 0.9772
0.0302 1.2121 200 0.0397 0.9420 0.9562 0.9860 0.9937 0.8783 0.9965 0.9870 0.8609 0.9781
0.0378 1.5152 250 0.0366 0.9447 0.9804 0.9856 0.9916 0.9655 0.9840 0.9872 0.8704 0.9765
0.232 1.8182 300 0.0278 0.9582 0.9810 0.9893 0.9894 0.9599 0.9938 0.9875 0.9026 0.9844
0.023 2.1212 350 0.0252 0.9630 0.9821 0.9905 0.9958 0.9595 0.9910 0.9895 0.9141 0.9852
0.0254 2.4242 400 0.0263 0.9626 0.9841 0.9901 0.9964 0.9675 0.9885 0.9887 0.9146 0.9846
0.0153 2.7273 450 0.0299 0.9613 0.9735 0.9906 0.9952 0.9290 0.9963 0.9904 0.9080 0.9855
0.0172 3.0303 500 0.0230 0.9645 0.9776 0.9913 0.9956 0.9417 0.9956 0.9917 0.9153 0.9864
0.0338 3.3333 550 0.0185 0.9723 0.9875 0.9928 0.9972 0.9733 0.9922 0.9913 0.9368 0.9889
0.0168 3.6364 600 0.0231 0.9679 0.9788 0.9922 0.9969 0.9438 0.9958 0.9921 0.9237 0.9878
0.0253 3.9394 650 0.0245 0.9664 0.9772 0.9918 0.9965 0.9388 0.9962 0.9920 0.9202 0.9869
0.0163 4.2424 700 0.0191 0.9689 0.9832 0.9923 0.9961 0.9592 0.9943 0.9917 0.9270 0.9881
0.0133 4.5455 750 0.0173 0.9745 0.9877 0.9932 0.9976 0.9728 0.9928 0.9913 0.9428 0.9895
0.0133 4.8485 800 0.0171 0.9742 0.9876 0.9934 0.9965 0.9721 0.9942 0.9921 0.9405 0.9901
0.0362 5.1515 850 0.0178 0.9725 0.9866 0.9931 0.9973 0.9692 0.9934 0.9918 0.9360 0.9897
0.0142 5.4545 900 0.0208 0.9679 0.9888 0.9919 0.9961 0.9797 0.9904 0.9919 0.9244 0.9874
0.0111 5.7576 950 0.0149 0.9772 0.9882 0.9941 0.9964 0.9727 0.9956 0.9924 0.9478 0.9915
0.0184 6.0606 1000 0.0165 0.9737 0.9822 0.9934 0.9977 0.9525 0.9963 0.9915 0.9388 0.9909
0.0181 6.3636 1050 0.0157 0.9759 0.9853 0.9938 0.9973 0.9628 0.9959 0.9924 0.9443 0.9909
0.0138 6.6667 1100 0.0143 0.9781 0.9907 0.9943 0.9966 0.9811 0.9945 0.9926 0.9501 0.9917
0.0287 6.9697 1150 0.0161 0.9747 0.9875 0.9934 0.9976 0.9714 0.9935 0.9920 0.9420 0.9900
0.0144 7.2727 1200 0.0149 0.9774 0.9894 0.9940 0.9974 0.9771 0.9938 0.9920 0.9493 0.9909
0.012 7.5758 1250 0.0139 0.9783 0.9906 0.9943 0.9971 0.9805 0.9942 0.9929 0.9506 0.9915
0.0098 7.8788 1300 0.0134 0.9793 0.9901 0.9945 0.9976 0.9782 0.9945 0.9927 0.9533 0.9918
0.0105 8.1818 1350 0.0182 0.9780 0.9895 0.9942 0.9971 0.9768 0.9946 0.9926 0.9500 0.9913
0.014 8.4848 1400 0.0141 0.9784 0.9896 0.9943 0.9969 0.9769 0.9948 0.9924 0.9512 0.9916
0.0117 8.7879 1450 0.0154 0.9767 0.9911 0.9938 0.9968 0.9834 0.9930 0.9917 0.9477 0.9908
0.0153 9.0909 1500 0.0143 0.9789 0.9908 0.9945 0.9964 0.9810 0.9951 0.9930 0.9518 0.9919

Framework versions

  • Transformers 4.41.2
  • Pytorch 2.0.1+cu117
  • Datasets 2.19.2
  • Tokenizers 0.19.1