[Doc]: about jetson AGX Orin,yolov8l OBB model predict time

[jaffe-fly]

for a single img

all predict time: 0.0109 秒
all predict time: 0.0109 秒
all predict time: 0.0109 秒

for batch=6

all predict time: 0.0496 秒
all predict time: 0.0511 秒
all predict time: 0.0480 秒
all predict time: 0.0524 秒
all predict time: 0.0519 秒

test code

def tensorrt_yolo_infer():
    print("======start tensorrt_yolo_infer ===========")
    MODEL_PATH = "/opt/yolo/test/yolobox_speed/output/yolov8l-obb-640-55.engine"
    option = InferOption()
    option.enable_swap_rb()
    # option.enable_performance_report()

    model = OBBModel(MODEL_PATH, option)

    rio_list = []
    for img_path in image_cut_path:
        image = cv2.imread(img_path)
        # image = cv2.resize(image, (640, 480))
        # print(image.shape)
        rio_list.append(image)

    for _ in range(50):
        time.sleep(3)
        sti = time.time()
        results = model.predict(rio_list)
        print(f"all predict time: {time.time() - sti:.4f} 秒")
        # model.performance_report()

use trtyolo get onnx and trtexec get engine
why cant get about 3.6ms with yolo11l-OBB in doc?

[laugh12321]

@jaffe-fly, where does the 3.6ms latency figure you mentioned come from? Is it from the figure below? However, it’s important to note that the figure tests the latency of YOLO detection, not the performance of OBB (Oriented Bounding Box). Additionally, the hardware used in the test was an RTX 2080Ti, not a Jetson AGX Orin. There could be significant differences in performance and latency between the two.

[jaffe-fly]

yes, share my data for others, jetson AGX Orin seems to be a bit inferior to the RTX
by my test in 3090, same batch use pt model to predict about 0.55ms,equal to above batch predict time
if use this repo to predict maybe less time-consuming

[jaffe-fly]

share RTX3090 test result

for a single img

batch=1 all predict time: 0.004 秒
batch=1 all predict time: 0.004 秒
batch=1 all predict time: 0.005 秒
batch=1 all predict time: 0.004 秒

for batch=6

batch=6 all predict time: 0.015 秒
batch=6 all predict time: 0.015 秒
batch=6 all predict time: 0.015 秒
batch=6 all predict time: 0.015 秒