search

lewes6369 / TensorRT-Yolov3

TensorRT for Yolov3
MIT License
489 stars 165 forks source link

wrong output box with model trained by myself #67

Closed zerollzeng closed 4 years ago

zerollzeng commented 4 years ago

hi @lewes6369, recently I test your model with my own car detection model, It has only 1 class, and the output seems strange. as you can see, this is official darknet result, it works fine. image but when I run it with tensorrt, some output is wrong image see detection at the bottom, I also test it with other image, and get some fimiliar error image

image

I print some information for this images image and this is network output points in cropped image image as you can see, the point below the image is strange, I think it should be the car which is at the middle bottom of the image. do you have any idea about this?

zerollzeng commented 4 years ago

after check the code, I think it should be something wrong with model. I'll check it in details.