* dice loss * format code, add docstring and calculate denominator without valid_mask * minor change * restore * add metafile * add manifest.in and add config at setup.py * add requirements * modify manifest * modify manifest * Update MANIFEST.in * add metafile * add metadata * fix typo * Update metafile.yml * Update metafile.yml * minor change * Update metafile.yml * add subfix * fix mmshow * add more metafile * add config to model_zoo * fix bug * Update mminstall.txt * [fix] Add models * [Fix] Add collections * [fix] Modify collection name * [Fix] Set datasets to unet metafile * [Fix] Modify collection names * complement inference time |
||
---|---|---|
.. | ||
README.md | ||
deeplabv3_r101-d8_512x1024_80k_fp16_cityscapes.py | ||
deeplabv3plus_r101-d8_512x1024_80k_fp16_cityscapes.py | ||
fcn_r101-d8_512x1024_80k_fp16_cityscapes.py | ||
metafile.yml | ||
pspnet_r101-d8_512x1024_80k_fp16_cityscapes.py |
README.md
Mixed Precision Training
Introduction
@article{micikevicius2017mixed,
title={Mixed precision training},
author={Micikevicius, Paulius and Narang, Sharan and Alben, Jonah and Diamos, Gregory and Elsen, Erich and Garcia, David and Ginsburg, Boris and Houston, Michael and Kuchaiev, Oleksii and Venkatesh, Ganesh and others},
journal={arXiv preprint arXiv:1710.03740},
year={2017}
}
Results and models
Cityscapes
Method | Backbone | Crop Size | Lr schd | Mem (GB) | Inf time (fps) | mIoU | mIoU(ms+flip) | config | download |
---|---|---|---|---|---|---|---|---|---|
FCN | R-101-D8 | 512x1024 | 80000 | 5.37 | 8.64 | 76.80 | - | config | model | log |
PSPNet | R-101-D8 | 512x1024 | 80000 | 5.34 | 8.77 | 79.46 | - | config | model | log |
DeepLabV3 | R-101-D8 | 512x1024 | 80000 | 5.75 | 3.86 | 80.48 | - | config | model | log |
DeepLabV3+ | R-101-D8 | 512x1024 | 80000 | 6.35 | 7.87 | 80.46 | - | config | model | log |