Use the NGC CLI to download:
Copied!
Resource
U-Net allows for seamless segmentation of 2D images, with high accuracy and performance.
Use the NGC CLI to download:
Copied!
Changelog
February 2021
- Updated training and inference performance with V100 and A100 results
- Refactored the example scripts
June 2020
- Updated training and inference accuracy with A100 results
- Updated training and inference performance with A100 results
February 2020
- Initial release
Known issues
- Training on 8 GPUs with FP32 and XLA using batch size 8 may sometimes cause out-of-memory errors.
- For TensorFlow 2.0 the training performance using AMP and XLA is around 30% lower than reported here. The issue was solved in TensorFlow 2.1.