What it does

This repository collects implementations of standard deep learning algorithms—LeNet-5, AlexNet, VGG, GoogLeNet, and Inception variants—written in both PaddlePaddle and TensorFlow. It accompanies a Chinese-language blog series that walks through theory, then code, then small practical projects like digit recognition, license plate recognition, and spam detection. The author started the series in 2017 and switched to PaddlePaddle’s Fluid 1.4 API after May 2018.

The interesting bit

The dual-framework structure is the point. For learners coming from TensorFlow tutorials, seeing the same network in PaddlePaddle’s Fluid API makes the framework-specific boilerplate visible rather than invisible. The repo also includes some practical topics—transfer learning, VisualDL visualization, data preprocessing—that often get skipped in algorithm-centric tutorials.

Key highlights

• Side-by-side implementations: same architecture, two frameworks
• Covers CNN classics from LeNet through Inception v4
• Includes applied projects: digit recognition, license plate OCR, spam classification
• Accompanied by detailed Chinese blog posts with theory explanations
• Uses older but stable versions: Python 2.7, PaddlePaddle Fluid 1.4

Caveats

• Python 2.7 and older framework versions; not maintained for current PaddlePaddle or TensorFlow 2.x
• Last blog update was June 2018; some links and APIs may be stale
• README is sparse on actual repository structure—browse the code to see what’s implemented

Verdict

Worth a look if you’re a Chinese-speaking beginner trying to understand how framework APIs differ, or if you specifically need PaddlePaddle examples from the Fluid era. Skip it if you want production-ready, modern code or TensorFlow 2/Keras patterns.