Rectangular training for v3 about yolov6 HOT 5 OPEN

Hi @KaranBhuva22 , you could git pull the latest code from main branch, and download the corresponding pretrained weights from here under Assets. Then you can fine-tune YOLOv6 models with rectangular training parameter. Otherwise, you could also modify the code referring to code from main branch.

from yolov6.

Hello @Chilicyy , could you please clarify why the --rect parameter is missing in infer.py on the main branch? Additionally, how would one go about running inference if the model has been trained in rectangular mode?

from yolov6.

Hi @KaranBhuva22 , you can specify img-size when running inference, for example --img-size 640 512.

from yolov6.

Hi @Chilicyy , in rectangular training mode, the image sizes vary from one another. Here's an illustration:

Image 1:
Original Image Shape: (418, 1368, 3)
Resized Image Shape: (224, 640, 3)

Image 2:
Original Image Shape: (2000, 1285, 3)
Resized Image Shape: (640, 413, 3)

As demonstrated, the width and height of each image may differ. This raises a concern regarding how to specify the --img-size parameter during inference. Kindly correct me if my understanding is incorrect!

from yolov6.