SigNet implementation in Pytorch
Original paper: https://arxiv.org/pdf/1707.02131.pdf
Note: I'm not the author of the paper. I'm just curious about the network and topic that I want to reimplement myself in Pytorch
cd data
./cedar.sh
This is used to split dataset to train/test partitions
python3 prepare_data.py
python3 train.py
Dataset | Accuracy (%) |
---|---|
CEDAR | 100 |
- Init bias = 0 as paper proposed
- Add normalization (e.g. divide by standard deviation)
- GPDS 300 dataset
- GPDS Synthetic Signature Corpus dataset
- Bengali dataset
- Hindi dataset
- (Maybe author's code) Keras implementation https://github.com/sounakdey/SigNet