/segment-anything-tensorrt

Primary LanguageJupyter NotebookMIT LicenseMIT

Overview

中文-windows

download translated onnx model

tutorials

vim_h trt model tested on GPU 3060 Ti, tests ok.
tutorials.ipynb 
tutorials_vim_h.ipynb

how to export vim_h model

how_to_export_vim_h_model.ipynb

# Divide the embeding model into 2 parts, named as part1 and part2 
# The decoder model remains unchanged
# export_embedding_model_part_1()
# export_embedding_model_part_2()
# export_sam_h_model()
# onnx_model_example2()

FP32 or FP16

you can not set FP16 value in export.h, if gpu have enough mems.