Convert Pytorch pretrain -> TensoRT engine directly for CRAFT (Character-Region Awareness For Text detection)
- Convert CRAFT Text detection pretrain Pytorch model into TensorRT engine directly, without ONNX step between
- CRAFT: (forked from https://github.com/clovaai/CRAFT-pytorch) Official Pytorch implementation of CRAFT text detector | Paper | Pretrained Model | Supplementary
- Using torch2trt_dynamic from https://github.com/grimoire/torch2trt_dynamic (branch of https://github.com/NVIDIA-AI-IOT/torch2trt with dynamic shapes support)
Implementation of inference pipeline using Tensor RT for CRAFT text detector. Two modules included:
- Convert pretrain Pytorch -> ONNX -> TensorRT
- Inference using Tensor RT
Note: This repo is about converting steps to finally get Tensor RT engine, and inference on the engine. More related repo about Tensor RT inference, check out:
- Advance inference pipeline using NVIDIA Triton Server (https://github.com/k9ele7en/triton-tensorrt-CRAFT-pytorch)
- Convenient converter from Pytorch to Tensor RT directly, without ONNX bridge step (https://github.com/k9ele7en/torch2tensorRT-dynamic-CRAFT-pytorch).
k9ele7en. Give 1 star if you find some value in this repo.
Thank you.
[MIT License] A short, permissive software license. Basically, you can do whatever you want as long as you include the original copyright and license notice in any copy of the software/source.
7 Aug, 2021: Init repo, converter run success. Run infer by ONNX success. Run infer by RT engine return wrong output.
$ pip install -r requirements.txt
Check details at ./README_Env.md
| Model name | Used datasets | Languages | Purpose | Model Link |
|---|---|---|---|---|
| General | SynthText, IC13, IC17 | Eng + MLT | For general purpose | Click |
| IC15 | SynthText, IC15 | Eng | For IC15 only | Click |
| LinkRefiner | CTW1500 | - | Used with the General Model | Click |
sh prepare.sh
$ cd converters
$ python pth2onnx.py
$ python onnx2trt.py
$ python infer_trt.py
$ python infer_onnx.py