yolov8obb data save and real-time detect by map50

[zhang990327]

When using yolo obb for detection, the detected data cannot be saved according to time, confidence and detection names.
a: If i set save-txt=true, it will be saved according to x1y1xyx3y3x4y4.not saved according to time, confidence and detection names

b: when i use chatgpt wrote coding, After saving, there will be a problem such as AttributeError: 'NoneType' object has no attribute 'data'.

I want to know when detecting, I want to use the same evaluation index as yolov4, that is, use map50 for detection instead of map50-95 for detection.
How to solve these two problems?

[UltralyticsAssistant]

👋 Hello @zhang990327, thank you for your interest in Ultralytics 🚀! We recommend checking out the Ultralytics Docs for more guidance on using YOLO models effectively, including saving outputs and evaluation methods. Below are some tailored suggestions for your queries:

---

Saving Detected Data

If you're encountering an issue with saving detected data, please ensure you've followed the appropriate configurations in the detection script. It may also be helpful to include a 🐛 minimum reproducible example of your code so we can better understand and debug any issues.

Also, depending on your use case, double-check the command-line options in the Docs to configure your desired behavior, such as saving data to specific formats or directories.

---

mAP Evaluation Index

To modify the evaluation metric from mAP50-95 to mAP50, you can use configuration settings or post-processing to specify which metric to prioritize. If there’s a specific command or function where you're seeing unexpected behavior, please share the detailed code or logs, as this would help us pinpoint the precise cause.

---

Automation Script Challenges

Regarding the NoneType error you mentioned when using code generated by ChatGPT, it sounds like the object in question might be improperly initialized. Sharing the relevant code snippet and the exact stack trace would greatly help with debugging.

---

Recommendations

1. Upgrade Ultralytics and Dependencies
   Ensure you are using the latest version of the ultralytics package and its dependencies in a compatible environment:

   pip install -U ultralytics

   The latest release often includes bug fixes and improvements that might resolve your issue.

2. Try Pre-configured Environments
   Run YOLO in an up-to-date, pre-configured environment for seamless testing and development. Examples include:

   • Colab Notebook:
   • Kaggle Notebook:
   • Paperspace Gradient:

3. Usage Tips for Customizations
   Refer to the Model Training Tips Guide to optimize your evaluation and detection configuration.

4. Community Support
   Join our lively Discord 🎧 for real-time assistance or engage in in-depth discussions on Discourse. You can also share ideas and feedback with us on our Subreddit!

---

Status Check

To ensure this isn't an external issue, verify the repository status:

If the CI badge is green, all Ultralytics CI tests are currently passing across different environments and tasks.

---

⚡ This is an automated response to help streamline your issue resolution. Rest assured, an Ultralytics engineer will assist you shortly! 😊

[zhang990327]

import os
import cv2
import sys
import numpy as np
import time
from ultralytics import YOLO
from ctypes import cast, POINTER, c_ubyte, c_int
from MvCameraControl_class import *
from datetime import datetime, timedelta
import pytz

YOLOv8 OBB 모델 로드

yolov8_obb_model = YOLO('yolov8n-obb.pt')

def get_video_save_path(base_path):
"""
비디오 파일 저장 경로를 생성합니다.
"""
seoul_tz = pytz.timezone('Asia/Seoul')
current_time = datetime.now(seoul_tz)

year_folder = os.path.join(base_path, current_time.strftime("%Y"))
if not os.path.exists(year_folder):
    os.makedirs(year_folder)

sub_folder = os.path.join(year_folder, current_time.strftime("%m_%d_%H"))
if not os.path.exists(sub_folder):
    os.makedirs(sub_folder)

file_name = current_time.strftime("%Y_%m_%d_%H_%M_%S") + ".mp4"
file_path = os.path.join(sub_folder, file_name)

return file_path, sub_folder

def detect_with_yolov8_obb(image):
"""
YOLOv8-OBB를 사용하여 검색 결과를 반환합니다.
"""
if len(image.shape) == 2:
image = cv2.cvtColor(image, cv2.COLOR_GRAY2RGB)

results = yolov8_obb_model(image, conf=0.25, save=False, save_txt=False)

if results and len(results) > 0:
    annotated_frame = results[0].plot()
else:
    annotated_frame = image

return annotated_frame

def grab_image_from_camera(cam):
"""
카메라에서 이미지 프레임을 가져옵니다.
"""
stOutFrame = MV_FRAME_OUT_INFO_EX()
width = c_int()
height = c_int()

cam.MV_CC_GetIntValue("Width", width)
cam.MV_CC_GetIntValue("Height", height)

data_size = width.value * height.value
data_buf = (c_ubyte * data_size)()

ret = cam.MV_CC_GetOneFrameTimeout(data_buf, data_size, stOutFrame, 1000)
if ret != 0:
    print(f"이미지 가져오기 실패! 오류 코드: 0x{ret:x}")
    return None

try:
    if len(data_buf) == stOutFrame.nWidth * stOutFrame.nHeight:
        image = np.frombuffer(data_buf, dtype=np.uint8).reshape((stOutFrame.nHeight, stOutFrame.nWidth))
    else:
        print("데이터 크기 불일치")
        return None
except ValueError as e:
    print(f"이미지 재구성 실패: {e}")
    return None

return image

def access_get_image_with_yolov8_obb(cam):
"""
카메라에서 이미지를 가져와 YOLOv8-OBB로 탐지한 후 저장합니다.
"""
base_path = "D:\yolov8"
frame_rate = 13.5
target_width, target_height = 640, 480

video_path, folder_path = get_video_save_path(base_path)
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
out = cv2.VideoWriter(video_path, fourcc, frame_rate, (target_width, target_height), isColor=True)

if not out.isOpened():
    print(f"비디오 파일을 열 수 없습니다: {video_path}")
    return

next_hour = (datetime.now(pytz.timezone('Asia/Seoul')) + timedelta(hours=1)).replace(minute=0, second=0, microsecond=0)

while True:
    frame = grab_image_from_camera(cam)
    if frame is None:
        print("카메라에서 이미지를 가져올 수 없습니다.")
        break

    annotated_frame = detect_with_yolov8_obb(cv2.cvtColor(frame, cv2.COLOR_GRAY2RGB))
    resized_frame = cv2.resize(annotated_frame, (target_width, target_height))
    out.write(resized_frame)

    cv2.imshow('YOLOv8-OBB Detection', resized_frame)

    current_time = datetime.now(pytz.timezone('Asia/Seoul'))
    if current_time >= next_hour:
        out.release()
        print(f"비디오 저장 완료: {video_path}")

        video_path, folder_path = get_video_save_path(base_path)
        out = cv2.VideoWriter(video_path, fourcc, frame_rate, (target_width, target_height), isColor=True)

        if not out.isOpened():
            print(f"비디오 파일을 열 수 없습니다: {video_path}")
            break

        next_hour = (current_time + timedelta(hours=1)).replace(minute=0, second=0, microsecond=0)

    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

out.release()
cam.MV_CC_StopGrabbing()
cam.MV_CC_CloseDevice()
cam.MV_CC_DestroyHandle()
cv2.destroyAllWindows()

def main():
deviceList = MV_CC_DEVICE_INFO_LIST()
cam = MvCamera()

ret = cam.MV_CC_EnumDevices(MV_GIGE_DEVICE | MV_USB_DEVICE, deviceList)
if ret != 0:
    print(f"장치 나열 실패! 오류 코드: 0x{ret:x}")
    sys.exit()

if deviceList.nDeviceNum == 0:
    print("장치를 찾을 수 없습니다!")
    sys.exit()

stDeviceList = cast(deviceList.pDeviceInfo[0], POINTER(MV_CC_DEVICE_INFO)).contents
ret = cam.MV_CC_CreateHandle(stDeviceList)
if ret != 0:
    print(f"핸들 생성 실패! 오류 코드: 0x{ret:x}")
    sys.exit()

ret = cam.MV_CC_OpenDevice(MV_ACCESS_Exclusive, 0)
if ret != 0:
    print(f"장치 열기 실패! 오류 코드: 0x{ret:x}")
    sys.exit()

ret = cam.MV_CC_SetFloatValue("AcquisitionFrameRate", 13.5)
if ret != 0:
    print(f"FPS 설정 실패! 오류 코드: 0x{ret:x}")
    sys.exit()

ret = cam.MV_CC_StartGrabbing()
if ret != 0:
    print(f"이미지 수집 시작 실패! 오류 코드: 0x{ret:x}")
    sys.exit()

access_get_image_with_yolov8_obb(cam)

if name == "main":
main()

[glenn-jocher]

@zhang990327 thank you for sharing your implementation! To ensure detection results are saved with time, confidence, and detection names, you can modify the save_txt function in your code to include these details. For evaluating with mAP50 instead of mAP50-95, you can use the metrics.box.map50 property after validation. Refer to the Ultralytics metrics documentation for further guidance. Let us know if additional clarification is needed!

[zhang990327]

spbo:
thanks for your explian. but after changing coding,i got a error like:AttributeError: 'NoneType' object has no attribute 'data',i donnot how to solve it .
and the second question. Is it accurate to use iou=50 when testing?

[glenn-jocher]

@zhang990327 the AttributeError: 'NoneType' object has no attribute 'data' error suggests an issue with how your results or objects are being accessed. Ensure that your input data and model outputs are correctly handled and not None during processing. For your second question, you can set iou=0.5 during evaluation by specifying it in your validation command or script. For example, in Python, you can use model.val(iou=0.5) or in CLI, yolo val model=your_model.pt iou=0.5. Let me know if you need further clarification!

[zhang990327]

Regarding the source code of yolov8obb, which source file is related to the content it outputs?

[glenn-jocher]

@zhang990327 the source file related to YOLOv8 OBB output content is ultralytics/models/yolo/obb/val.py, specifically the save_one_txt function for saving detection results. You can review its implementation for customization.