- clone this repo.
- From a terminal, run
python app.py
then open your browser and go to 127.0.0.1:5000
- Flask
- Python
- Tested on Google Chrome.
The UI enables reading an image with its openpose detection output (.json). Once the image file and JSON file is loaded, the tool plots a point each keyjoint.
- To see more information about each keypoint, you can use hover with your mouse.
- You can select a keypoint and move it around.
- You see the selected keypoint ID and its confidence at the status box.
- You can change the confidence of a selected keypoint by pressing ("+" or "c") keys to increase it by 20% every time you press. You can also decrease a keypoint confidence by pressing ("-" or "d") keys to decrease a keypoint confidence by 20%.
- View options can choose to hide some keypoints.
- These keypoints are divided as they are in openpose.
- There are skeleton keypoints, face keypoints, and (left, right) hand keypoints.
You can also search for a keypoint in using its id. A keypoint id is defined by the person_id, category id, keypoint number.
- The person id can be found by hover the mouse on another keypoint of that person.
- The keypoint category id is one of [0: skeleton, 1: face, 2: left hand, 3: right hand].
- The keypoint id can be recovered from the reference images in the top.
- You can also delete a keypoint by clicking on it then pressing DEL or Backspace.
- Deleting a keypoint sets it to (0, 0) position with confidence 0%.
- You can also add a whole new skeleton that does not exist in the uploaded json file.
- You can do this by clicking the button add new person. This will generate a list of all possible keypoints for the new person, however, all of them will be initialized on 0,0 with 0% confidance.
- You can search for the keypoint id (personID_catID_keypointID) that you want to place, this will bring them to the center.
- After you finish correcting your keypoints, you can save a new JSON file with the new keypoints.
- The file will be named with a postfix "_corrected.json"
Please cite this repo in case of using it.
- search by id
- add zoom preview on the image while hover.
- add option to control the confidence score of selected keypoints.
- add option to add a new skeleton (that does not exist in the openpose file)
- add option to multi-select keypoints by dragging on the image.
- add possibility to read other types of keypoint files (e.g. COCO)
- read from a folder to pass through multiple images