Connection Reduction of DenseNet for Image Recognition

Connection Reduction of DenseNet for Image Recognition

Figure

Citation

If you find ThreshNet useful in your research, please consider citing:

@article{ju2022connection,
 title={Connection Reduction of DenseNet for Image Recognition},
 author={Rui-Yang Ju, Jen-Shiun Chiang, Chih-Chia Chen, Yu-Shian Lin},
 conference={International Symposium on Intelligent Signal Processing and Communication Systems (ISPACS)},
 year={2022}
 }

Contents

1. Usage
2. Results
3. Requirements
4. Config
5. References

Usage

python3 main.py

optional arguments:

--lr                default=1e-3    learning rate
--epoch             default=200     number of epochs tp train for
--trainBatchSize    default=100     training batch size
--testBatchSize     default=100     test batch size

Results

Name	C10 GPU Time (ms)	C10 Error (%)	SVHN GPU Time (ms)	SVHN Error (%)	FLOPs (G)	MAdd (G)	Memory (MB)	#Params (M)	MenR+W (MB)
Baseline43	72.83	14.00	72.64	5.95	509.38	1.02	6.08	2.17	25.93
ShortNet1-43	61.17	13.59	58.97	5.65	374.00	0.75	4.60	1.59	18.92
ShortNet2-43	52.48	14.09	50.61	5.48	256.44	0.51	4.00	0.97	13.74
Baseline53	94.25	13.38	92.11	5.92	783.20	1.56	7.37	3.15	35.46
ShortNet1-53	71.19	13.36	69.57	5.63	536.76	1.07	5.41	2.16	24.56
ShortNet2-53	58.14	14.08	55.34	6.59	334.76	0.67	4.37	1.20	16.05

* GPU Time is the inference time per 100 images on NVIDIA RTX 3050

Requirements

• Python 3.6+
• Pytorch 0.4.0+
• Pandas 0.23.4+
• NumPy 1.14.3+

Config

Optimizer

• Adam Optimizer

Learning Rate

• 1e-3 for [1,74] epochs
  
• 5e-4 for [75,149] epochs
  
• 2.5e-4 for [150,200) epochs
  

References

GitHub

* [torchstat](https://github.com/Swall0w/torchstat) * [pytorch-cifar10](https://github.com/soapisnotfat/pytorch-cifar10)