A Tensorflow SqueezeNet implementation of the research https://arxiv.org/abs/1602.07360
SqueezeNet is a convolutional neural network with 80.3% top-5 accuracy on the ImageNet dataset. This architecture aims to signifanctly reduce the size of the model.
- Building the network
- Training the network on imagenet
- Currently published pre-trained weights cannot be used in TensorFlow, the goal of this project is to produce easy-to-use save files.
- Neural Style Transfer (http://arxiv.org/abs/1508.06576) on mobile devices
- Running on mobile devices in reasonable time without sacraficing quality too much like other feed-forward approaches.
