Learn Deep Learning for Image Processing with One Tutorial, Two Frameworks

You've probably felt the frustration: you find a great deep learning tutorial, but it only covers PyTorch. Or you're a TensorFlow person, and every example you stumble across is written for the other framework. It's a common problem, and it's exactly what this project—a comprehensive tutorial series from WZMIAOMIAO—sets out to solve.

The repository, deep-learning-for-image-processing, is a structured, video-based course that walks you through the major neural network architectures used in image processing. The twist? Every single network is taught twice: once with PyTorch and once with TensorFlow (using its Keras module). You learn the theory, then you build it in whichever framework you prefer—or both, if you're feeling ambitious.

What It Does

This is a curriculum, plain and simple. The author compiled their graduate school research into a series of video lessons, each following a consistent three-step flow: first, they explain the network's structure and its innovations; second, they show you how to build and train it in PyTorch; third, they repeat the process using TensorFlow's Keras API.

The scope is impressive. The image classification section alone covers a who's who of modern architectures: LeNet, AlexNet, VGG, GoogLeNet, ResNet, ResNeXt, MobileNet V1/V2/V3, ShuffleNet, EfficientNet V1/V2, RepVGG, Vision Transformer, Swin Transformer, ConvNeXt, and MobileViT. Beyond classification, the repository also branches into object detection with tutorials on Faster-RCNN and FPN, though the README notes the curriculum is still growing.

All the lecture slides are stored in a course_ppt folder, so you can follow along with the visuals even if you prefer reading over watching. The videos themselves are hosted on Bilibili, a major Chinese video platform, and each section in the README links directly to the relevant video.

Why It's Cool

The dual-framework approach is the killer feature here, and it's rare to see it done this systematically.

• Learn the concepts once, apply them anywhere. The theoretical knowledge transfers. But seeing the same architecture implemented in both PyTorch and TensorFlow is genuinely illuminating—you start to recognize which parts are framework-specific and which are core deep learning concepts. That's a level of understanding you don't get from a single-framework tutorial.

• It's a structured path through the field's history. You're not jumping around randomly. The curriculum follows the chronological evolution of image classification networks, from LeNet's early days through the CNN boom (VGG, ResNet) to the modern transformer-based architectures like ViT and Swin Transformer. It's a well-organized way to build your mental model of how the field developed.

• The format respects your time. Each network gets dedicated videos for the theory, the PyTorch implementation, and the TensorFlow implementation. You can skip straight to the framework you care about. The theory videos are separate from the coding ones, so you're never forced to sit through implementation details you don't need.

• The author practices what they teach. This isn't a collection of slides from a random online course. It's a summary of the author's actual graduate research, which means the material has been battle-tested in real academic work. That lends a credibility you don't always get from tutorial content.

• It stays current. The README explicitly states the curriculum will expand as the author learns new material. The inclusion of recent architectures like ConvNeXt and MobileViT shows this isn't stuck in 2018—it's actively tracking where the field is heading.

How to Try It

Getting started is straightforward. Head over to the repository and browse the README's table of contents. Each entry links directly to the corresponding videos on Bilibili.

Here's a suggested path:

1. Pick a starting point. If you're new to deep learning, begin with LeNet—it's the classic entry point. If you're more experienced, jump to an architecture you're curious about, like Vision Transformer or ConvNeXt.
2. Watch the theory video first. Each section starts with a network explanation video. This gives you the architectural context you'll need for the implementation.
3. Choose your framework. After the theory, watch either the PyTorch or TensorFlow implementation video. If you're learning both frameworks, watching them back-to-back is a great exercise in understanding framework differences.
4. Grab the slides. Download the relevant PPT files from the course_ppt folder to have the visuals in front of you while you code.

Note that the videos are hosted on Bilibili, so you'll want to make sure you can access the platform. The content itself is in Chinese, so if you don't speak the language, you may need to rely on subtitles or the code in the videos.

Final Thoughts

This repository is best for developers who want a structured, comprehensive introduction to image classification and object detection architectures, especially if you're trying to become fluent in both PyTorch and TensorFlow. The dual-framework format is genuinely unique and forces a deeper understanding than you'd get from a single-framework course.

That said, if you're looking for English-language content or prefer text-based tutorials over video, this might not be the perfect fit. And the Bilibili hosting means you'll need to work with that platform's interface.

But if you're serious about image processing and want to see how the field's major networks are built across the two dominant frameworks, this is a resource worth bookmarking. And since the author plans to keep expanding the curriculum, there's a good chance this will only get more valuable over time.

Follow @githubprojects for more developer tools and open source projects.