TensorFlow Technology Development and Practical Applications: Deep Learning Overview, TensorFlow Introduction, and Fashion Design Use Cases

The talk, delivered by Zheng Zeyu (Co‑founder & CEO of ZhiYi Technology) at the AI Pioneer Conference, covered the development of deep learning, an introduction to TensorFlow, and its practical deployment in the fashion design domain.

It began with a brief history of deep learning, highlighting milestones such as AlphaGo and AlphaZero, explaining the motivation for using AI in tasks lacking explicit procedural rules, and describing the basic architecture of neural networks (input, hidden, and output layers) with examples ranging from sentiment classification to image recognition.

The presentation then introduced TensorFlow, noting its origins in Google’s DistBelief system, its transition to a GPU‑enabled framework, rapid iteration cycles, and its dominance over other deep‑learning libraries since 2015, making it a recommended entry point for newcomers.

Next, the speaker explored opportunities and challenges of applying deep learning to fashion design, emphasizing the scarcity of AI solutions in this sector, the difficulty of aggregating multi‑source, heterogeneous, and massive data, and ZhiYi Technology’s approach of leveraging image recognition and personalized recommendation to assist designers.

The labeling problem was discussed in detail: fashion tags are sparse and often require fine‑grained detail (e.g., specific sleeve styles). By involving professional designers in the annotation process, the company achieved high‑accuracy labeling (93‑95% recognition) and built a dedicated tagging system to mitigate sparsity and missing labels.

For practical implementation, a two‑stage Faster R‑CNN based pipeline was described. The first stage detects garments, while the second applies style‑specific filters and clustering to support two search scenarios: whole‑image outfit matching and fine‑grained similar‑item retrieval.

Finally, the article provided a brief biography of Zheng Zeyu, noting his background at Google, authorship of a leading TensorFlow book, and his research contributions to machine learning conferences, followed by information about the DataFun community.