PaddleClas

PaddleClas is a PaddlePaddle toolkit for developers and researchers building image classification and recognition applications on their own hardware. It covers fixed-category classifiers and image matching for large or frequently changing catalogs.

• Image retrieval: PP-ShiTuV2 matches images against a reference collection using object detection, image features, and vector search, with tools to manage the collection for tasks such as product, logo, and vehicle recognition.
• Lightweight classification: PULC provides models for people and vehicle attributes, helmet classification, text orientation, and image clarity assessment.
• Model choices: The library includes MobileNetV4 and FasterNet classifiers, CLIP-based recognition and multi-label models, plus MobileFaceNet and ResNet50_face for face recognition.
• Custom training: You can adapt pretrained models with data augmentation and knowledge distillation, then compress models for deployment; PaddleX adds a graphical interface for model use and customization.

PaddleClas is open source under Apache 2.0 and supports deployment on your own server or on-device, with Python and C++ inference support and Paddle2ONNX export. PP-LCNet targets Intel CPUs, PP-LCNetV2 works with OpenVINO, and PP-HGNet targets GPUs. PaddleX supports model development on NVIDIA GPUs and hardware from Kunlunxin, Ascend, Cambricon, and Hygon. PP-ShiTuV2 also has an Android demo.