Train yolov3, yolov5 or yolov8 with image resolution of 224 x224 for small object detection

YOLOv5 🚀
1

Hello everyone,

I am trying to detect small objects which are around 30 pixels in an HD image. I cropped image with size of 640x640 to train with yolo5 but the results are not good. The reason is the object of interest is very small and dataset is also not big.

I wanted to crop image to 224x224 or 332x332 (i.e. any samller size) and want to train the yolo architecture. I want to fed this size as it is and do not want the algorithm to resize it to its input resolution for example 640x640 as it distorts the object shape.

Is there any way I can do it? I need guidance.

2

I really need help on this. Can anyone tell me that if the image size is 224x224, YOLOv5 will resize it to 640x640 or it takes original size of image (224x224) as input?

3

From what i know, the Ultralytics library accepts 3 types of image size as the parameter which are 640, 1280, and 1920. I don’t know if we’re able to put custom size as the input.