Pix2Seq is a multi-task computer vision framework that reformulates tasks like object detection as sequence generation problems. The codebase supports both autoregressive decoding and diffusion-based generative models (including FitTransformer, Bit Diffusion, and RIN). Built in TensorFlow 2 with efficient TPU/GPU support, it provides pretrained checkpoints for tasks like object detection on datasets like Objects365 and provides Colab notebooks for inference.