Training Techniques | Nesterov Accelerated Gradient, Weight Decay, Linear Warmup With Cosine Annealing, Label Smoothing |
---|---|
Architecture | ReLU6, Convolution, Batch Normalization, Residual Connection, Dropout |
ID | rexnet_100 |
SHOW MORE |
Training Techniques | Nesterov Accelerated Gradient, Weight Decay, Linear Warmup With Cosine Annealing, Label Smoothing |
---|---|
Architecture | ReLU6, Convolution, Batch Normalization, Residual Connection, Dropout |
ID | rexnet_130 |
SHOW MORE |
Training Techniques | Nesterov Accelerated Gradient, Weight Decay, Linear Warmup With Cosine Annealing, Label Smoothing |
---|---|
Architecture | ReLU6, Convolution, Batch Normalization, Residual Connection, Dropout |
ID | rexnet_150 |
SHOW MORE |
Training Techniques | Nesterov Accelerated Gradient, Weight Decay, Linear Warmup With Cosine Annealing, Label Smoothing |
---|---|
Architecture | ReLU6, Convolution, Batch Normalization, Residual Connection, Dropout |
ID | rexnet_200 |
SHOW MORE |
Rank Expansion Networks (ReXNets) follow a set of new design principles for designing bottlenecks in image classification models. Authors refine each layer by 1) expanding the input channel size of the convolution layer and 2) replacing the ReLU6s.
To load a pretrained model:
import timm
m = timm.create_model('rexnet_100', pretrained=True)
m.eval()
Replace the model name with the variant you want to use, e.g. rexnet_100
. 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.
@misc{han2020rexnet,
title={ReXNet: Diminishing Representational Bottleneck on Convolutional Neural Network},
author={Dongyoon Han and Sangdoo Yun and Byeongho Heo and YoungJoon Yoo},
year={2020},
eprint={2007.00992},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
MODEL | TOP 1 ACCURACY | TOP 5 ACCURACY |
---|---|---|
rexnet_200 | 81.63% | 95.67% |
rexnet_150 | 80.31% | 95.16% |
rexnet_130 | 79.49% | 94.67% |
rexnet_100 | 77.86% | 93.88% |