"Hardcore" labelme shows the label in the picture

This is a niche blog that most friends may not be able to use, but it is definitely a good medicine for those who need it!Hope it will be useful to those in need!

Although labelme is mainly used for data labeling, when the labelled data set is checked, checks whether the label corresponding to the label box is correct. This work is not commonly used, but it is not used when it is used.It is very inconvenient. Each picture needs to be clicked one by one, and then check whether the corresponding label is correct, which is very time-consuming and labor-intensive.

Here, on the basis of labelme, bloggers display the corresponding labels together, which makes it no longer necessary to click to view them one by one, but to scan them directly, which is very convenient and can be increased by at least 10 times.

Modified code to upload to viatorsun.blog.csdn.net
Just in case, it is recommended that friends only include app.py and widgets/canvas.py , widgets/label_dialog.py file replacement, not recommended to replace __main__.py

A total of two files need to be modified. The specific operations are as follows:

1, app.py

In the def loadFile(self, filename=None): function, add the following two lines of code at about 1533 lines of code

2, widgets/canvas.py

The script imports several packages of PyQt5from PyQt5.QtGui import QPainter,QFont,QColor, in def __init__(self, *args, **kwargs):Add the attribute self._shapes = [] to record the annotation information, and finally add the following information in def paintEvent(self, event):

3. Display of newly added annotations

4. Modify the size of the Edit label dialog box

In the widgets/label_dialog.py file, add the following statement after line 48 [super(LabelDialog, self).__init__(parent)], and friends can modify the width according to their own conditionsHigh ratio

from qtpy.QtCore import QSizeself.resize(QSize(300, 500))

This is it

Q&A

The project is relatively simple, there may be hidden bugs, please forgive me!!!

1. Tags are not displayed in real time: they are displayed only when the corresponding data is read for the first time when the file is opened.The current target data needs to be saved in a JSON file before it can be displayed, which can be displayed through the [Next Image/Prev Image] operation
2. According to the above operation, there may be cases where the displayed label will be covered by the callout box. Here, you can move the drawn code to the front of p.end(), so that the correspondingLabel on top shows
   
3. Because the images of the dataset are different, the color of the paint brush may not have strong contrast, so if there is a small partner whose label is not particularly obvious, you can modify the color of the paint brush by yourselfp.setFont(QColor( , , )) just