Qt_Autoannotater Instructions:

1. Required packages: (more comprehensive & detailed package in requirements.txt)

   • pytorch
   • PyQt6
   • PyQt5
   • pandas
   • PIL
   • numpy
   • random
   • os
   • sys
   • cv2
   • yaml
   • torchvision
   • ultralytics
   • transformers
   • copy
   • gc
   • sklearn
   • tkinter
   • argparse
   • shapely
   • ast
   • sys
   • maybe more let me know

2. Prepare needed files:

   • A folder of images
   • Label file
   • *** prepare a txt file with each line being a class of tools/objects ***

3. Once running:

   • click Rregular Annotate' for mannual annotation and 'Auto Annotate' for semi-automated annotation
   • click 'load label classes' to select the class txt file created in step 2
   • click 'open file' to select the folder where the images are stored
   • click 'OK' when it says the key frame selection is initiated and wait
   • mannualy annotate the selected images after finishing one image click 'D' to save labels
   • after all selected images are annotated click 'auto annotate'
   • click 'yes' if you have correctly annotated these images and ok to wait for model training and predicting
   • the programm will show the next set of selected images with predicted bouding boxed on them, correct them then click 'D' to complete the reviewing for each image
   • once the set is reviewed, click auto annotate again, it will run less training epochs from here
   • once the all the images are labeled and reviewed, clicking on 'auto annotate' will tell you auto annotate is completed and show you all images
   • You Label file should already be updated, you can try visualizer to see if this is actually correct

4. Background:

   • *** the images are fixed at 1920 * 1080, if yours is not this ratio change it here : self.image_width = 1920 self.image_height = 1080 ***
   • wait means that the window will freeze, please try not to do anything at that moment
   • the labels are saved in YOLO txt file format for all labeled images in the image folder and they are also updated in your Label file # Additionally, while annotation the working process and results can be observed in the pulled Qt_Autoannotater folder's Model folder
   • The program enables Manual select number of frames to annotate by checking that box, this is because the auto selection will become very frequent and only a small number of images will be selected for editing, not recommended but you can turn it on to accelerated the annotation process when confident and needed
   • *** Change box thickness and label font size if needed
     • rect_item.setPen(QPen(color, 10)) is the box thickness
     • font_size = 40 is the for label txt size
     • And try not use key "A" for adjust These all may cause the image to shrink in certain circumstances

Annotation commands

1. Adding bounding box: click key 1~9 to select your object class in the same order of the class txt file you created then left click to drag and drop
2. Deleting bounding box: hover you pointer on a box and click 'E' or 'Backspace'
3. Adjusting bounding box (not recommended): hover pointed on a box and click 'A' drage the squares shown to adjust box
4. Moving bounding box: right click and hold a box to move to where ever you want
5. Clear all labels: click 'C' to remove all boxes on the current image
6. Copy and paste from the previous image (not recommended): click 'V'
7. Go to next image and save: click 'D'
8. Go to previous image without saving: click 'S'
9. Go to 10 imgae later and save (not recommended): click 'F'