This is an lib which target at inference model using TensorRT and model from onnx format. Nowdays massive model are trained using pytorch and it's easy to export into onnx format, with the benefit of huge acceleration by TensorRT, combines onnx && TensorRT would be a very important in production.
Currently, we support models as:
All you need is install TensorRT 6.0 and install on:
# move it into ~/TensorRT (or soft link)
./build_all.sh
Refinenet Segmentation:
the refinenet body part segmentation onnx model can be download from 链接: https://pan.baidu.com/s/1zpUAKY2CwVYaKi9QnWi66Q 提取码: aium . The training framework from here, you can training your down dataset and replace the example onnx model to yours, it will works as you expected.
./examples/demo_refinenet no_have_serialize_txt float32 save_serialize_name here_your_video_file_name_or_cam here_your_onnxmodel_name
./trt_engine_refinenetseg refinenet_seg.trt /media/fagangjin/wd/permanent/datasets/TestVideos/elonmask.mp4
2. **RetinaFace Landmark Detection**:
this model already included under the `models` folder. you can convert it to trt first, then load directly into infer engine for inference. Make sure you have TensorRT 6 installed, and you shuold get onnx model from here: http://manaai.cn/aicodes_detail3.html?id=46
./build_all.sh ./build/examples/trt_engine_retinaface ../models/retinaface.trt ./images/0--Parade_0_Parade_marchingband_1_657.jpg
## Build
to build the lib and examples, you gonna need TensorRT download, we find TensorRT from `~/tensorrt` by default. Then run:
./build_all.sh
## Copyright
All rights reserved by Fagang Jin, codes released under Apache License.
此处可能存在不合适展示的内容,页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容,可点击提交进行申诉,我们将尽快为您处理。