Python version == 3.7.5
git clone https://github.com/ZachKLYeh/Annotation_Selector.git
cd Annotation_Selector
python main.py
One way of annotation is to train the model on small annotated dataset
Then utilize the model to predict some unlabeled images
Then pick the mispredicted images for annotation
The application is to fasten this behavior
By helping annotator to view the prediction of the model and select wrong prediction for futher annotation
The application will read input folder and get image/annotation pairs. According to file name.
Xml format and txt(yolo) format are supported.
The input folder structure shoud be like this:(xml format visualization)
|-- input foler
| |-- img1.jpg
| |-- img1.xml
| ˋ-- ...
Or like this:(txt format visualization)
|-- input foler
| |-- img1.jpg
| |-- img1.txt
| ˋ-- ...
Or like this:(txt format visualization)
|-- input foler
| |--labels
| |-- img1.txt
| |-- img2.txt
| ˋ-- ...
| |-- img1.jpg
| |-- img2.jpg
| ˋ-- ...
-
Press A for previous image
-
Press D for next image
-
Press ctrl + mouse scroll to zoom in/zoom out
-
Press C to select/unselect an image
-
Press E to display/hide object classes
When you pressed the "move selected" button
The application will move all the selected image/annotation pair to output folder.
Note: We are only moving xml format annotation
Then you can edit them via LabelImg