Home

Welcome to LFD wiki!

LFD is an One-Stage Anchor-Free Multi-Class detector for generic purpose. LFFD, LFD's previous version, has been proved to be highly practical for many applications and is popular among developers. Inspired by the success of LFFD, we release LFD with many new features:

• implemented using PyTorch (vs MXNet for LFFD)
• support multi-class detection (vs single-class of LFFD)
• higher precision & lower inference latency
• easy to create diverse network structures and train from scratch
• a flexible framework for easy re-development
• long-term support

In the future, we hope LFD can be as popular as YOLO series.

---

In this wiki, we deliver the following main content:

• framework design
• details of LFD
• how to apply LFD to practical applications

---

LFD is not just an object detection method, it is a way to deeply understand the detection and know how to devise a desired model for real-world applications. We hope that you can get more than you want here!