Deep neural network for traffic sign recognition systems: An analysis of spatial transformers and stochastic optimisation methods

Pytorch Implementation of Deep neural network for traffic sign recognition systems: An analysis of spatial transformers and stochastic optimisation methods

• ÁlvaroArcos-GarcíaJuan A.Álvarez-GarcíaLuis M.Soria-Morillo* Deep neural network for traffic sign recognition systems: An analysisof spatial transformers and stochastic optimisation methods

link1 link2

LCN Implementation is taken from https://github.com/dibyadas/Visualize-Normalizations

Notes:

• ASGD Works best among all optimizers for me for Learning Rate : 10^-2
• Class imbalance is removed prior to training by duplicating the data
• Learning Rate Decay worsenes the performance
• Data Augmentation, in general, decreases performance although Spatial Transformer model ensures augmentation isn't a problem.
• Architecture is changed slighlty from the original set of layers
• Currently Gaussian filter is kept constant for LCN, where as ideally it should be chosed at random during run-time

Neural Net gives output of 6 neurons necessary for Affine transformation (translation, cropping, rotation, scaling, and skewing) and uses grid generator and sampling as inbuilt Pytorch commands

• Max Jaderberg Karen Simonyan Andrew Zisserman Koray Kavukcuoglu
  Spatial Transformer Networks (https://arxiv.org/pdf/1506.02025.pdf )

Citing

BibTeX

@misc{pathaktraffic,
  author = {Priyank Pathak},
  title = {PyTorch Deep Neural Network for Traffic Sign Recognition Systems},
  year = {2018},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/ppriyank/Deep-neural-network-for-traffic-sign-recognition-systems}}
}