BR-IDL/PaddleViT

Implements 15+ transformer and MLP-based vision models (ViT, DeiT, Swin, VOLO, BEiT, etc.) with modular architectures that support image classification, object detection, semantic segmentation, and GANs. Built on PaddlePaddle 2.1+, it provides pretrained weights, unified training/validation pipelines, data augmentation utilities, and distributed training via DDP with mixed-precision support. Each model is exposed as a standalone module for rapid prototyping and fine-tuning on custom datasets.

1,241 stars. No commits in the last 6 months.