Output format of YOLOv4?

marcoslucianops / DeepStream-Yolo

NVIDIA DeepStream SDK 7.0 / 6.4 / 6.3 / 6.2 / 6.1.1 / 6.1 / 6.0.1 / 6.0 / 5.1 implementation for YOLO models

MIT License

1.4k stars 345 forks source link

Output format of YOLOv4? #382

Open aidevmin opened 1 year ago

aidevmin commented 1 year ago

Thanks @marcoslucianops for the amazing repo. I follow your guide for Darknet YOLOv4 model https://github.com/marcoslucianops/DeepStream-Yolo/tree/68f762d5bdeae7ac3458529bfe6fed72714336ca. I want to evaluate my engine model. I see in the in the raw output of engine model (before postprocessing) has bboxes, score and class. One raw bbox has 4 values and I am confusing about these values. Can you confirm:

Which format is for 4 values? (x1, y1, w, h or x1, y1, x2, y2 or x_center, y_center, w, h)
Values are absolute or relative to original image size (or input size of model)?

marcoslucianops commented 1 year ago

The output of the Darknet (parse-bbox-func-name=NvDsInferParseYolo) is:

In the engine file: [x_center, y_center, w, h] and the values are absolute in a range of 0 - input size of the model (640 for example).
In the deepstream code: you get the [left, top, width, height] from the obj_meta and the values are absolute in a range of 0 - streammux size (1920 for x and 1080 for y for example).