When using the model in production video, the bounding box x,y is in a different location then in the sample deploy screen. Have use the exact same video from production in the deploy screen; the deploy screen works fine in production video BB different location.
The video stream is 1980 x 1080. Is it because the model images are down sized?
The top image is the stream, with red line indicating the x/y of the model BB.
tks
Hi @bonjih, it may have something to do with the Deploy Tab downsizing images prior to sending them for inference.
I have a code example here for image resizing prior to inference: { "error": [ "Error opening image. Check that you are sending a properly formatted image." ] } - #2 by Mohamed
Additionally, is your model trained with images from this environment, and with the 1920x1080 video frames?
Hi Mohamed,
Yes, when I print out prediction, the width and height of the frame sent to the API for inference is 1920x1080, and I am using 1920x1080 for streaming.
I have tried to down size (using cv2.resize())) prior, with no success, not sure what to down size to.
tks for your reply.
Can you add Roboflow Support to the workspace so we can diagnose this issue fully?
added, tks Mohamed
