Simple annotation tool for bounding boxes in images that also enables pre-annotations (for example from a object detection model) to be loaded and corrected. No classes are required and also enables editing of image captions, allowing for creating annotations for visual grounding models.
The only prerequisites for runnning the tool are python and flask. Furthermore, the tool requires a hosted image library. To host a library on localhost simply run python -m http.server 8080 in the directory containing your images.
app.py expects an INPUT_FILE and an OUTPUT_FILE (both variables in app.py), which can be the same file. The input file must be a JSON file which contains a list of dictionaries where each list entry represents the pre-annotations for a single image. Each entry must follow the following key-value strcuture:
boxes: list of bounding boxes inxywhformat, relative to the image size (i.e. between 0 and 1). May be empty list.phrases: list of bounding box labels, matching in index to theboxeslist. May be empty list.caption: string containing description of the image. May be empty string.image_path: URI of the image which lives within the hosted image library.
static/js/script.js contains a global variable imageUrl which points to the root URL of the hosted image library.
Run python app.py to launch the tool.