Open luameows opened 3 years ago
also the resized img is not in the center of the blank, but top left of the blank, so the code should be changed.
rsz_img.copyTo(flt_img(cv::Rect(IMAGE_WIDTH/2 - rsz_img.cols/2, IMAGE_HEIGHT/2 - rsz_img.rows/2, rsz_img.cols, rsz_img.rows)));
resize style can change, but test results show RGB order is better than BGR order, so i use RGB order @luameows
In pytorch, the model is train with BGR img order, however in tensorrt, the input is RGB order.![image](https://user-images.githubusercontent.com/30432086/104433997-ad56d780-55c5-11eb-8438-f92b8dcb6811.png)