Image shape not correct while using detect.py

[kenrickfernandes]

I run detect.py on an mp4 file as follows:
python3 detect.py --source ./vid.mp4 --output /test/ml/ScaledYOLOv4/tta_results_32epoch/ --weights runs/exp5_yolov4-p5/weights/best.pt --img-size 896 --device 0 --conf-thres 0.75

But, while inferencing I see that the image shape is 512,896.
video 1/1 (1/8451) /test/ml/darknet_detection_training/test_vids/020096.mp4: 512x896 Done. (0.190s)

Is it supposed to be this way?
How do I change it to inference on (896,896) images?

[kenrickfernandes]

Closing this issue as I have found the reason. It is clearly stated as a comment in the dataloaders.
To understand more, look at : ultralytics/yolov3#232