yolact_edge
yolact_edge copied to clipboard
WisconsinAIVision
questions about how to deploy yolact_edge with openvino in c++?
@haotian-liu Hi, I'm a undergraduate in North University of China and I'm working on Robomaster competition which is held by DJI,I'm a member of the vision group which is responsibile for the autoaim function of the robot in the team.And I have used yolov5s to detect the armor with tensorrt in jetson nano but due to the low fps we finally turn to the Intel's NUC.Another problem is that when I use the ordinary object detection model like yolo series,the bbox can't fit the four corners very well which cause the coordinates of the armor solved through PnP is not so accurate.In that case,the team of SJTU first present four keypoint model based on yolov5 and they change the head and add a loss of the four corners so the detection results are much better because the model is directly regress the four corners of the armor.Besides,they also opensource a label tool which can autolabel the data with the pretrained model.Now I'm working on it based on yolov5-face and now I just need to write a script to convert the format of the dataset I have labeled with the tool produced by SJTU.However,I think if the instance segmantion is the same with the fourkrypoint model because the datasetof it also labeled by keypoint of the object.So now I want to try your model which can be faster on mobile device but I don't know whether the yolact_edge can be converted to onnx and then deployed with openviono in c++ because our code is c++,hope you can give me some advices:-)
the armor is like this and the bbox is the result of the four keypoint model:
Hey Hezhexi2002 !
It is possible to convert it to ONNX and work with c++, a good starting Point is here:
https://github.com/PINTO0309/PINTO_model_zoo https://github.com/PINTO0309/PINTO_model_zoo/tree/main/085_Yolact_Edge/30_Full_Converted_mbnv2_550x550 https://github.com/PINTO0309/PINTO_model_zoo/blob/main/085_Yolact_Edge/30_Full_Converted_mbnv2_550x550/yolact_test.cpp
Hey Hezhexi2002 !
It is possible to convert it to ONNX and work with c++, a good starting Point is here:
https://github.com/PINTO0309/PINTO_model_zoo https://github.com/PINTO0309/PINTO_model_zoo/tree/main/085_Yolact_Edge/30_Full_Converted_mbnv2_550x550 https://github.com/PINTO0309/PINTO_model_zoo/blob/main/085_Yolact_Edge/30_Full_Converted_mbnv2_550x550/yolact_test.cpp
Thank you for your reply，I have stared the repo PINTO0309 before.But I haven't try it，I will have a try soon and I will give you a feedback！
I have tried, but the ONNX conversion process for this repository was not easy.
Are you successful?
