FRTSearch: An End-to-End Detection and Characterization Tool for Fast Radio Transients

FRTSearch is an end-to-end detection and characterization tool for fast radio transients (FRTs) including Pulsars, Rotating Radio Transients (RRATs), and Fast Radio Bursts (FRBs) in radio astronomical observation data. This repository contains Python source code (.py) for the FRTSearch pipeline, including the Mask R-CNN-based detector and the IMPIC (Iterative Mask-based Parameter Inference and Calibration) algorithm. It supports PSRFITS (.fits) and Sigproc Filterbank (.fil) input formats with 1/2/4/8/32-bit data. Dependencies: Python 3.10+, PyTorch 2.0+, MMDetection 3.x, PRESTO, and additional packages listed in requirements.txt. Model weights are available at https://huggingface.co/waterfall109/FRTSearch/tree/main/models. This software is associated with the paper "FRTSearch: Unified Detection and Parameter Inference of Fast Radio Transients using Instance Segmentation" (Zhang et al. 2026). The training dataset CRAFTS-FRT is available at https://doi.org/10.57760/sciencedb.Fastro.00038.