I am developing a demo using deepstream 4.0. The pipeline consists of a detector and a classifier. The detector works fine. bboxes are correctly drawed on the video. But my classification model needs some custom preprocess, like resize and crop. I just don’t know where to add resize and crop operation.Please help.
For example, resolution of the input video is 1920x1080. A detected bbox is {left:1166,top:155,width:85,height:179}. I want to resize the bbox to 256x256 and then crop to 224x224.224x224 is the size of input tensor of my classifier. Any idea how to implement these?
It supports resize, but does not support crop now.
You can try to modify detector bbox result in “attach_metadata_detector”
left = 1166 * (224/256)
…