segformer-b0-winter
This model is a fine-tuned version of nvidia/mit-b0 on the johanhag/winter-test dataset. It achieves the following results on the evaluation set:
- Loss: 0.1441
- Mean Iou: 0.8861
- Mean Accuracy: 0.9456
- Overall Accuracy: 0.9660
- Accuracy Unlabeled: nan
- Accuracy Object: nan
- Accuracy Road: 0.9769
- Accuracy Side walk: 0.8930
- Accuracy Car: 0.9347
- Accuracy Pedestrian: nan
- Accuracy Other: 0.9779
- Iou Unlabeled: nan
- Iou Object: nan
- Iou Road: 0.9197
- Iou Side walk: 0.8250
- Iou Car: 0.8319
- Iou Pedestrian: nan
- Iou Other: 0.9678
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: 6e-05
- train_batch_size: 2
- eval_batch_size: 2
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 50
Training results
Training Loss | Epoch | Step | Validation Loss | Mean Iou | Mean Accuracy | Overall Accuracy | Accuracy Unlabeled | Accuracy Object | Accuracy Road | Accuracy Side walk | Accuracy Car | Accuracy Pedestrian | Accuracy Other | Iou Unlabeled | Iou Object | Iou Road | Iou Side walk | Iou Car | Iou Pedestrian | Iou Other |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
0.2207 | 4.0 | 20 | 0.2300 | 0.8576 | 0.9439 | 0.9583 | nan | nan | 0.9738 | 0.8767 | 0.9564 | nan | 0.9686 | nan | nan | 0.9018 | 0.8000 | 0.7670 | nan | 0.9617 |
0.1792 | 8.0 | 40 | 0.2126 | 0.8696 | 0.9457 | 0.9614 | nan | nan | 0.9768 | 0.8911 | 0.9444 | nan | 0.9706 | nan | nan | 0.9106 | 0.8122 | 0.7924 | nan | 0.9633 |
0.1527 | 12.0 | 60 | 0.1869 | 0.8769 | 0.9470 | 0.9634 | nan | nan | 0.9776 | 0.9023 | 0.9364 | nan | 0.9718 | nan | nan | 0.9180 | 0.8165 | 0.8085 | nan | 0.9647 |
0.1329 | 16.0 | 80 | 0.1787 | 0.8783 | 0.9429 | 0.9634 | nan | nan | 0.9772 | 0.8880 | 0.9314 | nan | 0.9749 | nan | nan | 0.9126 | 0.8117 | 0.8229 | nan | 0.9661 |
0.1746 | 20.0 | 100 | 0.1651 | 0.8864 | 0.9511 | 0.9668 | nan | nan | 0.9771 | 0.9126 | 0.9395 | nan | 0.9751 | nan | nan | 0.9258 | 0.8369 | 0.8157 | nan | 0.9671 |
0.1218 | 24.0 | 120 | 0.1652 | 0.8798 | 0.9444 | 0.9643 | nan | nan | 0.9791 | 0.8858 | 0.9370 | nan | 0.9757 | nan | nan | 0.9140 | 0.8156 | 0.8224 | nan | 0.9673 |
0.0816 | 28.0 | 140 | 0.1473 | 0.8921 | 0.9521 | 0.9684 | nan | nan | 0.9723 | 0.9199 | 0.9383 | nan | 0.9780 | nan | nan | 0.9299 | 0.8478 | 0.8231 | nan | 0.9676 |
0.0893 | 32.0 | 160 | 0.1490 | 0.8892 | 0.9502 | 0.9672 | nan | nan | 0.9749 | 0.9140 | 0.9354 | nan | 0.9766 | nan | nan | 0.9260 | 0.8384 | 0.825 | nan | 0.9673 |
0.0849 | 36.0 | 180 | 0.1517 | 0.8861 | 0.9476 | 0.9660 | nan | nan | 0.9791 | 0.8987 | 0.9367 | nan | 0.9760 | nan | nan | 0.9205 | 0.8258 | 0.8308 | nan | 0.9674 |
0.1625 | 40.0 | 200 | 0.1519 | 0.8843 | 0.9468 | 0.9654 | nan | nan | 0.9777 | 0.8938 | 0.9394 | nan | 0.9763 | nan | nan | 0.9176 | 0.8235 | 0.8289 | nan | 0.9673 |
0.1396 | 44.0 | 220 | 0.1500 | 0.8850 | 0.9476 | 0.9655 | nan | nan | 0.9791 | 0.8949 | 0.9408 | nan | 0.9757 | nan | nan | 0.9187 | 0.8223 | 0.8317 | nan | 0.9674 |
0.0931 | 48.0 | 240 | 0.1441 | 0.8861 | 0.9456 | 0.9660 | nan | nan | 0.9769 | 0.8930 | 0.9347 | nan | 0.9779 | nan | nan | 0.9197 | 0.8250 | 0.8319 | nan | 0.9678 |
Framework versions
- Transformers 4.35.2
- Pytorch 2.1.0+cu121
- Datasets 2.15.0
- Tokenizers 0.15.0
- Downloads last month
- 2
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.
Model tree for johanhag/segformer-b0-winter
Base model
nvidia/mit-b0
Finetuned
this model