metadata
license: other
base_model: nvidia/mit-b5
tags:
- vision
- image-segmentation
- generated_from_trainer
model-index:
- name: SegFormer_mit-b5_Clean-Set3_Augmented_Medium
results: []
SegFormer_mit-b5_Clean-Set3_Augmented_Medium
This model is a fine-tuned version of nvidia/mit-b5 on the Hasano20/Clean-Set3 dataset. It achieves the following results on the evaluation set:
- Loss: 0.0205
- Mean Iou: 0.9709
- Mean Accuracy: 0.9837
- Overall Accuracy: 0.9928
- Accuracy Background: 0.9979
- Accuracy Melt: 0.9593
- Accuracy Substrate: 0.9938
- Iou Background: 0.9927
- Iou Melt: 0.9315
- Iou Substrate: 0.9885
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: 20
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 |
---|---|---|---|---|---|---|---|---|---|---|---|---|
0.1481 | 0.3968 | 50 | 0.1951 | 0.7748 | 0.8972 | 0.9261 | 0.9808 | 0.8114 | 0.8996 | 0.9583 | 0.4859 | 0.8802 |
0.1364 | 0.7937 | 100 | 0.0745 | 0.8855 | 0.9155 | 0.9727 | 0.9933 | 0.7647 | 0.9884 | 0.9753 | 0.7206 | 0.9604 |
0.0776 | 1.1905 | 150 | 0.1238 | 0.8046 | 0.8351 | 0.9565 | 0.9922 | 0.5172 | 0.9960 | 0.9754 | 0.5059 | 0.9325 |
0.0851 | 1.5873 | 200 | 0.0878 | 0.8651 | 0.8901 | 0.9687 | 0.9923 | 0.6841 | 0.9940 | 0.9791 | 0.6665 | 0.9498 |
0.1893 | 1.9841 | 250 | 0.0602 | 0.9077 | 0.9628 | 0.9766 | 0.9958 | 0.9233 | 0.9693 | 0.9872 | 0.7755 | 0.9602 |
0.1236 | 2.3810 | 300 | 0.0643 | 0.9042 | 0.9455 | 0.9768 | 0.9971 | 0.8609 | 0.9783 | 0.9758 | 0.7669 | 0.9699 |
0.0809 | 2.7778 | 350 | 0.0387 | 0.9408 | 0.9699 | 0.9857 | 0.9964 | 0.9271 | 0.9862 | 0.9895 | 0.8562 | 0.9769 |
0.0357 | 3.1746 | 400 | 0.0364 | 0.9431 | 0.9701 | 0.9860 | 0.9948 | 0.9274 | 0.9880 | 0.9897 | 0.8629 | 0.9767 |
0.0408 | 3.5714 | 450 | 0.0424 | 0.9349 | 0.9815 | 0.9834 | 0.9936 | 0.9745 | 0.9765 | 0.9904 | 0.8434 | 0.9708 |
0.0973 | 3.9683 | 500 | 0.0541 | 0.9172 | 0.9798 | 0.9785 | 0.9941 | 0.9796 | 0.9655 | 0.9903 | 0.7997 | 0.9615 |
0.0274 | 4.3651 | 550 | 0.0256 | 0.9636 | 0.9830 | 0.9904 | 0.9967 | 0.9629 | 0.9894 | 0.9893 | 0.9170 | 0.9844 |
0.04 | 4.7619 | 600 | 0.0329 | 0.9482 | 0.9696 | 0.9877 | 0.9977 | 0.9210 | 0.9899 | 0.9888 | 0.8743 | 0.9815 |
0.0301 | 5.1587 | 650 | 0.0247 | 0.9609 | 0.9756 | 0.9905 | 0.9975 | 0.9361 | 0.9933 | 0.9917 | 0.9067 | 0.9844 |
0.0137 | 5.5556 | 700 | 0.0214 | 0.9667 | 0.9779 | 0.9919 | 0.9962 | 0.9414 | 0.9962 | 0.9924 | 0.9206 | 0.9870 |
0.019 | 5.9524 | 750 | 0.0243 | 0.9619 | 0.9817 | 0.9907 | 0.9959 | 0.9577 | 0.9915 | 0.9916 | 0.9090 | 0.9851 |
0.021 | 6.3492 | 800 | 0.0200 | 0.9678 | 0.9853 | 0.9920 | 0.9972 | 0.9668 | 0.9917 | 0.9926 | 0.9237 | 0.9871 |
0.0156 | 6.7460 | 850 | 0.0211 | 0.9689 | 0.9813 | 0.9924 | 0.9965 | 0.9520 | 0.9954 | 0.9926 | 0.9259 | 0.9880 |
0.0153 | 7.1429 | 900 | 0.0205 | 0.9685 | 0.9842 | 0.9923 | 0.9970 | 0.9626 | 0.9930 | 0.9930 | 0.9249 | 0.9876 |
0.0125 | 7.5397 | 950 | 0.0205 | 0.9709 | 0.9837 | 0.9928 | 0.9979 | 0.9593 | 0.9938 | 0.9927 | 0.9315 | 0.9885 |
Framework versions
- Transformers 4.41.2
- Pytorch 2.0.1+cu117
- Datasets 2.19.2
- Tokenizers 0.19.1