SVHNClassifier and SVHNClassifier-PyTorch
These are ecosystem siblings—parallel implementations of the same research paper in different deep learning frameworks (TensorFlow vs. PyTorch), allowing users to choose based on their preferred framework rather than requiring use of both together.
About SVHNClassifier
potterhsu/SVHNClassifier
A TensorFlow implementation of Multi-digit Number Recognition from Street View Imagery using Deep Convolutional Neural Networks (http://arxiv.org/pdf/1312.6082.pdf)
Implements variable-length digit sequence prediction using a multi-task CNN architecture with separate classifiers for digit count and individual digit positions, achieving 93.45% accuracy on the SVHN dataset. The pipeline converts raw image annotations to TFRecords format for efficient training and includes TensorBoard visualization and inference utilities for both dataset images and external photographs.
About SVHNClassifier-PyTorch
potterhsu/SVHNClassifier-PyTorch
A PyTorch implementation of Multi-digit Number Recognition from Street View Imagery using Deep Convolutional Neural Networks (http://arxiv.org/pdf/1312.6082.pdf)
Implements multi-task learning with separate digit length and individual digit classifiers (10 output classes per position, with class 10 representing "no digit"), achieving 95.65% accuracy on SVHN. The architecture uses convolutional feature extraction with multi-head classification branches trained jointly on the Street View House Numbers dataset. Includes LMDB data pipeline for efficient preprocessing and Visdom integration for training visualization, plus optional C++ inference backend.
Scores updated daily from GitHub, PyPI, and npm data. How scores work