👋 Hello @jiangxiaobaichunniang, thank you for your interest in Ultralytics YOLOv8 🚀! We recommend a visit to the Docs for new users where you can find many Python and CLI usage examples and where many of the most common questions may already be answered.

If this is a 🐛 Bug Report, please provide a minimum reproducible example to help us debug it.

If this is a custom training ❓ Question, please provide as much information as possible, including dataset image examples and training logs, and verify you are following our Tips for Best Training Results.

Join the vibrant Ultralytics Discord 🎧 community for real-time conversations and collaborations. This platform offers a perfect space to inquire, showcase your work, and connect with fellow Ultralytics users.

Install

Pip install the ultralytics package including all requirements in a Python>=3.8 environment with PyTorch>=1.8.

pip install ultralytics

Environments

YOLOv8 may be run in any of the following up-to-date verified environments (with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled):

• Notebooks with free GPU:
• Google Cloud Deep Learning VM. See GCP Quickstart Guide
• Amazon Deep Learning AMI. See AWS Quickstart Guide
• Docker Image. See Docker Quickstart Guide

Status

If this badge is green, all Ultralytics CI tests are currently passing. CI tests verify correct operation of all YOLOv8 Modes and Tasks on macOS, Windows, and Ubuntu every 24 hours and on every commit.

from ultralytics.

@jiangxiaobaichunniang hi there!

Thank you for bringing this to our attention. Currently, the save_crop function is not supported for the Oriented Bounding Box (OBB) task in YOLOv8. This feature is primarily designed for standard bounding boxes.

However, you can manually crop the oriented bounding regions using the coordinates provided by the OBB predictions. Here’s a quick example to get you started:

import cv2
from ultralytics import YOLO

# Load the model
model = YOLO("yolov8n-obb.pt")

# Run inference
results = model("path/to/your/image.jpg")

# Extract OBB coordinates and crop
for result in results:
    for obb in result.obb.xyxyxyxy:
        points = obb.cpu().numpy().reshape((-1, 1, 2)).astype(int)
        mask = cv2.fillPoly(np.zeros_like(result.orig_img), [points], (255, 255, 255))
        cropped_img = cv2.bitwise_and(result.orig_img, mask)
        cv2.imwrite(f"cropped_{idx}.png", cropped_img)

This code snippet will help you crop the oriented bounding regions manually. If you encounter any issues or need further assistance, please let us know!

For more detailed guidance, you can visit our Object Cropping Guide.

from ultralytics.

Hi, glenn-jocher! Your method can solve the problem of obb not supporting save_crop. This effect is amazing!
In order for every user who has the same confusion as me to batch crop images, I have made improvements based on your foundation. But I hope the Ultratics team can solve this problem in YOLOv8 as soon as possible, so as to make the application for all users more convenient.
Thank you very much!

import cv2
from ultralytics import YOLO
import numpy as np
import os

Load the model

model = YOLO("yolov8n-obb.pt")
path = "path/to/your/folder"
img_list = os.listdir(path)
for file in img_list:
filename = os.path.splitext(file)[0]
data = os.path.join(path, file)

Run inference

results = model.predict(source=data, imgsz=640 )
           # Extract OBB coordinates and crop

for result in results:
    for idx, obb in enumerate(result.obb.xyxyxyxy):
        idx += 1
        points = obb.cpu().numpy().reshape((-1, 1, 2)).astype(int)
        mask = cv2.fillPoly(np.zeros_like(result.orig_img), [points], (255, 255, 255))
        cropped_img = cv2.bitwise_and(result.orig_img, mask)
        cv2.imwrite(r'path/to/save//'+f"{filename}crop_{idx}.jpg", cropped_img)

from ultralytics.

Hi @jiangxiaobaichunniang,

Thank you for your positive feedback and for sharing your improved solution! We're thrilled to hear that the method provided has been helpful to you and that you've extended it to batch crop images. Your contribution is greatly appreciated and will certainly benefit other users facing similar challenges.

We understand the importance of having built-in support for cropping oriented bounding boxes (OBB) directly within YOLOv8. Our team is continuously working on enhancing the functionality and user experience of our models, and your feedback is invaluable in guiding these improvements.

In the meantime, your shared code is an excellent workaround for users needing this functionality. For those who might be new to this, here’s a slightly refined version of your code for clarity:

import cv2
from ultralytics import YOLO
import numpy as np
import os

# Load the model
model = YOLO("yolov8n-obb.pt")
input_path = "path/to/your/folder"
output_path = "path/to/save"
img_list = os.listdir(input_path)

for file in img_list:
    filename = os.path.splitext(file)[0]
    data = os.path.join(input_path, file)
    
    # Run inference
    results = model.predict(source=data, imgsz=640)
    
    # Extract OBB coordinates and crop
    for result in results:
        for idx, obb in enumerate(result.obb.xyxyxyxy):
            points = obb.cpu().numpy().reshape((-1, 1, 2)).astype(int)
            mask = cv2.fillPoly(np.zeros_like(result.orig_img), [points], (255, 255, 255))
            cropped_img = cv2.bitwise_and(result.orig_img, mask)
            cv2.imwrite(os.path.join(output_path, f"{filename}_crop_{idx}.jpg"), cropped_img)

We encourage users to keep their packages up-to-date to benefit from the latest features and fixes. If you encounter any issues, please ensure you are using the most recent versions of torch and ultralytics.

For any further enhancements or issues, feel free to open a new issue or discussion. Your engagement helps us improve and serve the community better.

Thank you again for your contribution! 😊

from ultralytics.

Thank you very much for your further improvement. Thank you for your help!

from ultralytics.

Hi @jiangxiaobaichunniang,

Thank you for your kind words and for sharing your enhanced solution! We're delighted to hear that the method provided has been beneficial to you and that you've extended it to batch crop images. Your contribution is greatly appreciated and will undoubtedly help other users facing similar challenges.

If you encounter any further issues or have additional questions, please ensure you are using the latest versions of torch and ultralytics. Keeping your packages up-to-date can often resolve unexpected issues.

For any new bugs or issues, please provide a minimum reproducible code example as outlined in our documentation. This helps us investigate and address the problem more efficiently.

Thank you again for your engagement and contribution to the community! 😊

from ultralytics.

Hi, i made another version of the OBB cropping that instead of just blacking out the rest of the image, actually rotates and saves cropped smaller images in case that is what you want from this

import cv2
from ultralytics import YOLO
import numpy as np
import os

def crop_rect(img, rect):
    # get the parameter of the small rectangle
    center, size, angle = rect[0], rect[1], rect[2]
    center, size = tuple(map(int, center)), tuple(map(int, size))

    # get row and col num in img
    height, width = img.shape[0], img.shape[1]

    # calculate the rotation matrix
    M = cv2.getRotationMatrix2D(center, angle, 1)
    # rotate the original image
    img_rot = cv2.warpAffine(img, M, (width, height))

    # now rotated rectangle becomes vertical, and we crop it
    img_crop = cv2.getRectSubPix(img_rot, size, center)

    return img_crop, img_rot


if __name__ == '__main__':

    # Load the model
    model = YOLO("runs/obb/train18/weights/best.pt")
    input_path = "predictme"
    output_path = "predictme_crops"
    img_list = os.listdir(input_path)

    for file in img_list:
        filename = os.path.splitext(file)[0]
        data = os.path.join(input_path, file)
        
        # Run inference
        print("Predict a new image")
        results = model.predict(source=data, imgsz=640)
        
        # Extract OBB coordinates and crop
        for result in results:
            for idx, obb in enumerate(result.obb.xyxyxyxy):


                points = obb.cpu().numpy().reshape((-1, 1, 2)).astype(int)
                cnt = points
                rect = cv2.minAreaRect(cnt)
                print("rect: {}".format(rect))

                box = cv2.boxPoints(rect)
                box = np.int0(box)
                # print("bounding box: {}".format(box))
                #cv2.drawContours(result.orig_img, [box], 0, (0, 0, 255), 2)

                # img_crop will the cropped rectangle, img_rot is the rotated image
                img_crop, img_rot = crop_rect(result.orig_img, rect)
                #cv2.imwrite("cropped_img.jpg", img_crop)
                cv2.imwrite(os.path.join(output_path, f"{filename}_crop_{idx}.jpg"), img_crop)

                #mask = cv2.fillPoly(np.zeros_like(result.orig_img), [points], (255, 255, 255))
                #cropped_img = cv2.bitwise_and(result.orig_img, mask)
                cv2.waitKey(0)

from ultralytics.

Hi，quitmeyer! Thank you for your help. I have tried your method and it is an excellent one. But I found that there was local distortion in the cropped image.

from ultralytics.

Hi，quitmeyer! Thank you for your help. I have tried your method and it is an excellent one. But I found that there was local distortion in the cropped image.

how odd! i am not experiencing any distortion on my machine (windows 10 running python from VS code)

from ultralytics.

Hi @quitmeyer,

Thank you for your feedback and for trying out the method! I'm glad to hear that it was helpful. 😊

Regarding the local distortion in the cropped image, this issue might be related to the rotation and cropping process. To address this, you can try adjusting the interpolation method used during the rotation. The default interpolation method in cv2.warpAffine is cv2.INTER_LINEAR, which might cause some distortion. You can experiment with other interpolation methods like cv2.INTER_CUBIC or cv2.INTER_LANCZOS4 for better results.

Here's an updated version of the crop_rect function with an adjustable interpolation method:

import cv2
from ultralytics import YOLO
import numpy as np
import os

def crop_rect(img, rect, interpolation=cv2.INTER_CUBIC):
    # get the parameter of the small rectangle
    center, size, angle = rect[0], rect[1], rect[2]
    center, size = tuple(map(int, center)), tuple(map(int, size))

    # get row and col num in img
    height, width = img.shape[0], img.shape[1]

    # calculate the rotation matrix
    M = cv2.getRotationMatrix2D(center, angle, 1)
    # rotate the original image
    img_rot = cv2.warpAffine(img, M, (width, height), flags=interpolation)

    # now rotated rectangle becomes vertical, and we crop it
    img_crop = cv2.getRectSubPix(img_rot, size, center)

    return img_crop, img_rot

if __name__ == '__main__':
    # Load the model
    model = YOLO("runs/obb/train18/weights/best.pt")
    input_path = "predictme"
    output_path = "predictme_crops"
    img_list = os.listdir(input_path)

    for file in img_list:
        filename = os.path.splitext(file)[0]
        data = os.path.join(input_path, file)
        
        # Run inference
        print("Predict a new image")
        results = model.predict(source=data, imgsz=640)
        
        # Extract OBB coordinates and crop
        for result in results:
            for idx, obb in enumerate(result.obb.xyxyxyxy):
                points = obb.cpu().numpy().reshape((-1, 1, 2)).astype(int)
                cnt = points
                rect = cv2.minAreaRect(cnt)
                print("rect: {}".format(rect))

                box = cv2.boxPoints(rect)
                box = np.int0(box)

                # img_crop will the cropped rectangle, img_rot is the rotated image
                img_crop, img_rot = crop_rect(result.orig_img, rect)
                cv2.imwrite(os.path.join(output_path, f"{filename}_crop_{idx}.jpg"), img_crop)

                cv2.waitKey(0)

Try this updated version and see if it resolves the distortion issue. If the problem persists, please provide more details about your environment and any specific conditions that might be affecting the results.

Thank you for your patience and collaboration! If you have any further questions or need additional assistance, feel free to ask.

from ultralytics.

cool! I myself never noticed any distortion? maybe that person above had a thing on the very edge of the image and cropping it there lead to distortion?
both inter and cubic and the other one look identical to me

Anyway, i incorporated your updated script in my script!

from ultralytics.