chriscodenation / densenet-with-aff-module

DenseNet-with-AFF-module

提出多尺度特徵融合之中文手寫字辨識模型，將DenseNet結合Attentional Feature Fusion，並搭配skip connection概念達到多層的特徵融合技術，使模型能取得更豐富的語意特徵，解決集成式模型資源不足的問題及降低模型複雜度，同時提升模型正確率。模型於ICDAR中文手寫辨識競賽的正確率從94.77%提升至97.23%，改善2.46%的正確率

Attentional feature fusion

ICDAR 2013 Chinese Handwriting Recognition Competition

What's in this repo so far

• code for training ICDAR dataset
• code for DenseNet with AFF module

Experiments

ICDAR2013 Test Datasets

Architecture	Top1 Accuracy(%)
Fujitsu[1]	94.77
IDSIAnn[2]	94.42
MCDNN[3]	95.79
ATR-CNN Voting[4]	96.06
EnsembleHCCRGoogLeNet-10[5]	96.74
DenseRAN[6]	96.66
GoogleNet-ResNet[7]	97.03
Google-ResNet+triplet loss[8]	97.03
ResNet+Center loss[9]	97.03
DenseNet+AFF(ours)	97.23

References

[1] Yin, F., Wang, Q. F., Zhang, X. Y., & Liu, C. L. (2013, August). ICDAR 2013 Chinese handwriting recognition competition. In 2013 12th international conference on document analysis and recognition (pp. 1464-1470). IEEE.

[2] Yin, F., Wang, Q. F., Zhang, X. Y., & Liu, C. L. (2013, August). ICDAR 2013 Chinese handwriting recognition competition. In 2013 12th international conference on document analysis and recognition (pp. 1464-1470). IEEE.

[3] Cireşan, D., & Meier, U. (2015, July). Multi-column deep neural networks for offline handwritten Chinese character classification. In 2015 international joint conference on neural networks (IJCNN) (pp. 1-6). IEEE.

[4] Wu, C., Fan, W., He, Y., Sun, J., & Naoi, S. (2014, September). Handwritten character recognition by alternately trained relaxation convolutional neural network. In 2014 14th International Conference on Frontiers in Handwriting Recognition (pp. 291-296). IEEE.

[5] Zhong, Z., Jin, L., & Xie, Z. (2015, August). High performance offline handwritten chinese character recognition using googlenet and directional feature maps. In 2015 13th international conference on document analysis and recognition (ICDAR) (pp. 846-850). IEEE.

[6] Wang, W., Zhang, J., Du, J., Wang, Z. R., & Zhu, Y. (2018, August). Denseran for offline handwritten chinese character recognition. In 2018 16th International Conference on Frontiers in Handwriting Recognition (ICFHR) (pp. 104-109). IEEE.

[7] Cheng, C., Zhang, X. Y., Shao, X. H., & Zhou, X. D. (2016, October). Handwritten Chinese character recognition by joint classification and similarity ranking. In 2016 15th International Conference on Frontiers in Handwriting Recognition (ICFHR) (pp. 507-511). IEEE.

[8] Cheng, C., Zhang, X. Y., Shao, X. H., & Zhou, X. D. (2016, October). Handwritten Chinese character recognition by joint classification and similarity ranking. In 2016 15th International Conference on Frontiers in Handwriting Recognition (ICFHR) (pp. 507-511). IEEE.

[9] Zhang, R., Wang, Q., & Lu, Y. (2017, November). Combination of ResNet and center loss based metric learning for handwritten Chinese character recognition. In 2017 14th IAPR International Conference on Document Analysis and Recognition (ICDAR) (Vol. 5, pp. 25-29). IEEE.