Training Techniques | Cosine Annealing, Random Erasing |
---|---|
Architecture | Convolution, ReLU, Batch Normalization, SelecSLS Block, Global Average Pooling, Dropout, Dense Connections |
ID | selecsls42b |
SHOW MORE |
Training Techniques | Cosine Annealing, Random Erasing |
---|---|
Architecture | Convolution, ReLU, Batch Normalization, SelecSLS Block, Global Average Pooling, Dropout, Dense Connections |
ID | selecsls60 |
SHOW MORE |
Training Techniques | Cosine Annealing, Random Erasing |
---|---|
Architecture | Convolution, ReLU, Batch Normalization, SelecSLS Block, Global Average Pooling, Dropout, Dense Connections |
ID | selecsls60b |
SHOW MORE |
SelecSLS uses novel selective long and short range skip connections to improve the information flow allowing for a drastically faster network without compromising accuracy.
To load a pretrained model:
import timm
m = timm.create_model('selecsls42b', pretrained=True)
m.eval()
Replace the model name with the variant you want to use, e.g. selecsls42b
. You can find the IDs in the model summaries at the top of this page.
You can follow the timm recipe scripts for training a new model afresh.
@article{Mehta_2020,
title={XNect},
volume={39},
ISSN={1557-7368},
url={http://dx.doi.org/10.1145/3386569.3392410},
DOI={10.1145/3386569.3392410},
number={4},
journal={ACM Transactions on Graphics},
publisher={Association for Computing Machinery (ACM)},
author={Mehta, Dushyant and Sotnychenko, Oleksandr and Mueller, Franziska and Xu, Weipeng and Elgharib, Mohamed and Fua, Pascal and Seidel, Hans-Peter and Rhodin, Helge and Pons-Moll, Gerard and Theobalt, Christian},
year={2020},
month={Jul}
}
BENCHMARK | MODEL | METRIC NAME | METRIC VALUE | GLOBAL RANK |
---|---|---|---|---|
ImageNet | selecsls60b | Top 1 Accuracy | 78.41% | # 156 |
Top 5 Accuracy | 94.18% | # 156 | ||
ImageNet | selecsls60 | Top 1 Accuracy | 77.99% | # 168 |
Top 5 Accuracy | 93.83% | # 168 | ||
ImageNet | selecsls42b | Top 1 Accuracy | 77.18% | # 192 |
Top 5 Accuracy | 93.39% | # 192 |