Simple Annotation KPT

This tool is a lightweight, no-AI version of the annotation pipeline.

It supports:

• extract: open a video (or image folder), split/export frames, and initialize dataset files.
• review: open a GUI to edit keypoints (frame navigation, add/delete/move points, category/point-id selection, save).

Outputs

For input <name>, outputs are written to:

• outputs/<name>/images/*
• outputs/<name>/center_points_masks_annotations.json
• outputs/<name>/keypoints.json

The JSON structure is kept compatible with auto_annotation_pipeline.

Desktop GUI (Windows)

Run without arguments to open the folder picker:

python E:\simple_annotation_kpt\main.py

Choose an input image folder (or video), an output root folder, and a frame step. The result directory is displayed as <output root>/<input name>. Click Extract first. Start annotation and Open result folder remain inactive until the result contains images and both valid annotation JSON files. Extraction runs in the background and the launcher displays its log. Existing result folders are protected against extraction overwriting annotations. To resume, select the same input and output root and click Start annotation.

The Python environment needs opencv-python, numpy, matplotlib, and Tkinter. The command-line commands below remain available.

Usage

1) Extract dataset (no AI)

python -u simple_annotation_kpt/main.py extract \
  --input /path/to/video_or_image_folder \
  --output outputs \
  --frame-step 1

2) Review keypoints

python -u simple_annotation_kpt/main.py review outputs/<name>

Reviewer Controls

• Q: quit
• A / D: previous / next frame
• S: save keypoints.json
• 1-5: set category
• Z/X/C/V: set point id to 1/2/3/4
• Left click: select point and drag to move
• Ctrl + Left click: add point using current category and point id, then clear selection.
• Auto next keypoint (enabled by default): after adding a point, keep the category unchanged and advance point id 1 -> 2 -> 3 -> 4 -> 1. Turn it off to add multiple points with the same point id, such as cells.
• Right click: delete nearest point

Annotation Guidelines

Please follow these rules when annotating:

• Tool tip: use Category 2 - Point 1.
• Tool root/base: use Category 2 - Point 2.
• Racket: use Category 1.
• For the racket, label the four points from the top-left corner in counterclockwise order as 0, 1, 2, 3.
• The model may swap pairs 0/1 and 2/3, so occasional swapped annotations are acceptable.
• Cells: use Category 3 - Point 1, and multiple points are allowed.