How to crop the detected object after training using YOLO?

[samahwaleed]

Search before asking

• I have searched the YOLOv5 issues and discussions and found no similar questions.

Question

I am using YOLO for model training. I used save-txt to save the results in txt files. I want to crop the detected object and get the pixel values for each detected object from the txt file. The txt file has labeled, bbox coordinates and score.

Any help would be highly appreciated!

Additional

No response

[glenn-jocher]

@samahwaleed 👋 Hello! Thanks for asking about cropping results with YOLOv5 🚀. Cropping bounding box detections can be useful for training classification models on box contents for example. This feature was added in PR #2827. You can crop detections using either detect.py or YOLOv5 PyTorch Hub:

detect.py

Crops will be saved under runs/detect/exp/crops, with a directory for each class detected.

python detect.py --save-crop

YOLOv5 PyTorch Hub

Crops will be saved under runs/detect/exp/crops if save=True, and also returned as a dictionary with crops as numpy arrays.

import torch

# Model
model = torch.hub.load('ultralytics/yolov5', 'yolov5s')  # or yolov5m, yolov5l, yolov5x, custom

# Images
img = 'https://ultralytics.com/images/zidane.jpg'  # or file, Path, PIL, OpenCV, numpy, list

# Inference
results = model(img)

# Results
crops = results.crop(save=True)  # or .show(), .save(), .print(), .pandas(), etc.

Good luck and let us know if you have any other questions!

[samahwaleed]

Thank you. but I want to crop the images using the results saved in txt file.
This is my code:

`import glob
import cv2 as cv
import matplotlib.pyplot as plt
import PIL
import numpy as np


path = glob.glob("/content/yolov5/runs/detect/exp/FrameVid1_185.JPG")


 for img in path:
    
     imgs = PIL.Image.open(img).convert('RGB')
     image_width, image_height = imgs.size
     imgArrays = np.array(imgs)
     X = (df.iloc[0:, 1]*image_width).astype(int)
     Y = (df.iloc[0:, 2]*image_height).astype(int)
     W = (df.iloc[0:, 3]*image_width).astype(int)
     H = (df.iloc[0:, 4]*image_height).astype(int)
     
     cropped_image = np.zeros(imgs.size)
     for i in range(len(X)):
       cropped_image[Y[i]: H[i], X[i]: W[i]] = imgArrays[Y[i]: H[i], X[i]: W[i]]
      
     cv2.imshow('Cropped Image', cropped_image)`

I got this error,

ValueError: could not broadcast input array from shape (0,0,3) into shape (0,0)

[glenn-jocher]

@samahwaleed crops are produced by the same text file predictions, but sure you can write your own script to do this as well.

[samahwaleed]

This is my code.

path = glob.glob("/content/yolov5/runs/detect/exp/*.jpg")
 cropped_images =[]

for img in path:
  
   imgs = PIL.Image.open(img).convert('RGB')
   image_width, image_height = imgs.size
   imgArrays = np.array(imgs)

   #image coordinates:
   #(left, right, top, bottom) = (xmin * im_width, xmax * im_width, ymin * im_height, ymax * im_height)
  
   X = (x_min*image_width)          #left
   Y = (y_min*image_height)         #top
   W = (x_max*image_width)          #right
   H = (y_max*image_height)         #bottom
   cropped_image = np.zeros((image_height, image_width, 3))
      
   for i in range(len(X)):
            x1, y1, w, h = X[i], Y[i], W[i], H[i]
            x_start = int(torch.round(x1 - (w/2)))
            y_start = int(torch.round(y1 - (h/2)))
            x_end = int(torch.round(x_start + w))
            y_end = int(torch.round(y_start + h))
            
            temp = imgArrays[y_start: y_end, x_start: x_end]
            cropped_images = temp

[github-actions]

👋 Hello, this issue has been automatically marked as stale because it has not had recent activity. Please note it will be closed if no further activity occurs.

Access additional YOLOv5 🚀 resources:

• Wiki – https://github.com/ultralytics/yolov5/wiki
• Tutorials – https://docs.ultralytics.com/yolov5
• Docs – https://docs.ultralytics.com

Access additional Ultralytics ⚡ resources:

• Ultralytics HUB – https://ultralytics.com/hub
• Vision API – https://ultralytics.com/yolov5
• About Us – https://ultralytics.com/about
• Join Our Team – https://ultralytics.com/work
• Contact Us – https://ultralytics.com/contact

Feel free to inform us of any other issues you discover or feature requests that come to mind in the future. Pull Requests (PRs) are also always welcomed!

Thank you for your contributions to YOLOv5 🚀 and Vision AI ⭐!

[Raymon360]

is there a method to cut-save multiple files in multiple folders?

[glenn-jocher]

@Raymon360 👋 Hello! Thanks for asking about cropping results with YOLOv5 🚀. Cropping bounding box detections can be useful for training classification models on box contents for example. This feature was added in PR #2827. You can crop detections using either detect.py or YOLOv5 PyTorch Hub:

detect.py

Crops will be saved under runs/detect/exp/crops, with a directory for each class detected.

python detect.py --save-crop

YOLOv5 PyTorch Hub

Crops will be saved under runs/detect/exp/crops if save=True, and also returned as a dictionary with crops as numpy arrays.

import torch

# Model
model = torch.hub.load('ultralytics/yolov5', 'yolov5s')  # or yolov5m, yolov5l, yolov5x, custom

# Images
img = 'https://ultralytics.com/images/zidane.jpg'  # or file, Path, PIL, OpenCV, numpy, list

# Inference
results = model(img)

# Results
crops = results.crop(save=True)  # or .show(), .save(), .print(), .pandas(), etc.

Good luck 🍀 and let us know if you have any other questions!

[Arsmart1]

can we only detect 'person' class and only save this class? Thank you!!!

[Laughing-q]

@Arsmart1

python detect.py --source data/images --weights yolov5s.pt --classes 0

this might help you, use --classes option to chose which class you want to detect.

[JJ-McClure]

I just want to chime in that cropping the image using the coordinates from the txt-file does not give the same image as using the save-crops function. The save-crops function result in a slightly bigger crop compared to just using the coordinates. This can be a very important distinction for small detections/images and sequential yolo detections

[varunpusarla]

Is there a better of way of doing this in YOLOv8 ?

[glenn-jocher]

@varunpusarla hello,

YOLOv8 is not an existing version of YOLO. The latest version is YOLOv5, which is currently maintained by Ultralytics. With YOLOv5, you can detect objects in images and videos using pre-trained models or custom trained models with your own dataset.

If you have any specific questions about implementation or functionality with YOLOv5, feel free to ask and we'll be happy to help.

Best regards.

[NawoyaSarah]

@glenn-jocher how can i maintain scale during cropping? because the generated crops vary in dimensions, yet i wound want to use the generated crops to measure object size. hence the dimensions need to be same scale

[glenn-jocher]

@Sarai256 you can maintain the scale during cropping by calculating the width and height of the crop based on a fixed aspect ratio or a fixed scale factor. For example, if you want to maintain a fixed aspect ratio (e.g., 1:1) during cropping, you can calculate the width and height of the crop based on the longer side of the bounding box. Alternatively, if you want to maintain a fixed scale factor (e.g., 0.5), you can scale the width and height of the crop based on the original width and height of the image. By applying these calculations, you can ensure that the dimensions of the generated crops are in the same scale as the original objects.

[NawoyaSarah]

@glenn-jocher , thank you for the reply. however I am not sure to which part of the code i should include the scale factor,
Example: python detect.py --save-crop..
Do i include the scale in that command? I am a beginner

[glenn-jocher]

@Sarai256 Thank you for reaching out! In order to maintain a specific scale during cropping, you would need to modify the code in the detect.py file.

The command python detect.py --save-crop does not have a built-in option to specify a scale factor. To achieve the desired scale, you would need to make changes in the code itself.

One way to implement scaling is by modifying the cropping logic in the detect.py file. Specifically, you can calculate the width and height of the crop based on the scale factor you want to maintain. You can then adjust the cropping coordinates accordingly.

If you are a beginner, I would recommend going through the detect.py code and understanding how the cropping is currently implemented. From there, you can make the necessary modifications to introduce the desired scaling functionality.

Feel free to ask any further questions if you need any clarification or assistance with the code. Happy coding!