Tool created in Python 3 with PyQt5 which annotates valid frames from a HDF5 dataset into a DB file created with sqlite3. Was used mainly to help discard unusable frames from CARLA created datasets where recorded data at certain points could be seen as undesirable. Can also be extended to be a more generalistic classification tool by working on the MainApplication class.
- Python 3
- PyQt5
- h5py
- cv2
- numpy
Clone this repo somewhere and install the prerequisites:
git clone https://github.com/AlanNaoto/Dataset_GUI_labelling_tool
pip3 install PyQt5 h5py opencv-python numpy
Inside the repo, run:
python3 main.py hdf5_file db_file
where:
- hdf5_file is the full path to the HDF5 dataset file
- db_file is the full path to the DB file. If it already exists, then it loads the last worked frame. If the DB file does not exists, then it creates a new one
If everything went right, the user interface should appear.
Available keystrokes are:
- Q - Changes the label marking to either "1" (good) or "0" (bad)
- Left and right arrows - Go back or advance one frame and saves the current label
The database file has one main table - frames_analysis with the columns timestamps and good_frame, where timestamps are the already covered timestamps from the HDF5 input file and good_frame an integer 0 or 1.
Icon 
made by andriy matviychuk from www.freeicons.io