SVHN-multi-digits recogniton

a tensorflow version implementation of Multi-digit Number Recognition from Street View Imagery using Deep Convolutional Neural Networks.

SVHN dataset

SVHN is obtained from house numbers in Google Street View images. The dataset can be download here(format1).

Training data samples

Requirements

python 3.6
tensorflow 1.4.0
numpy 1.15.0
matplotlib 2.0.0

Training details

we generate images with bounding boxes, and resize the images to 64×64. We then use the similar data augmentation which crops a 54×54 pixel image from a random location within the 64×64 pixel image in Goodfellow et al. (2013).

Run python convert_to_tfrecords.py, you can get three tfrecords files.
Run python main.py

Graph

Result

The recognition accuracy of this model is reached 96.04%. All qualitative and quantitative results are all exported to the svhn.log, you can print some other results to the logs if you are interested. You also can view results in tensorboard.

Run tensorboard --logdir=logs