This is the code for the paper Under Pressure: Learning-Based Analog Gauge Reading In The Wild by Maurits Reitsma, Julian Keller, Kenneth Blomqvist and Roland Siegwart.
Install Poetry
curl -sSL https://install.python-poetry.org | python3 -Install the project dependencies
poetry installEnter Poetry shell
poetry shellTo setup the conda environment to run all scripts follow the following instruction:
mkdir -p ~/miniconda3
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O ~/miniconda3/miniconda.sh
bash ~/miniconda3/miniconda.sh -b -u -p ~/miniconda3
rm -rf ~/miniconda3/miniconda.sh
~/miniconda3/bin/conda init bash
~/miniconda3/bin/conda init zshconda create --name gauge_reader python=3.8 -y
conda activate gauge_readerWe use torch version 2.0.0.
conda install pytorch==2.0.0 torchvision==0.15.0 torchaudio==2.0.0 -c pytorch -c nvidiaRefer to this page for installation https://mmocr.readthedocs.io/en/dev-1.x/get_started/install.html We use the version dev-1.x
pip install -U openmim
mim install mmengine==0.7.2
mim install mmcv==2.0.0
mim install mmdet==3.0.0
mim install mmocr==1.0.0We use the following versions: mmocr 1.0.0, mmdet 3.0.0, mmcv 2.0.0, mmengine 0.7.2. If for some reason the installation fails refer to open-mmlab/mmcv#2938. We found that it is essential that we have Pytorch version 2.0.0
We use ultralytics version 8.0.66
pip install ultralyticsWe use scikit-learn version 1.2.2
pip install -U scikit-learnThe pipeline script can be run with the following command:
python pipeline.py --detection_model path/to/detection_model --segmentation_model /path/to/segmentation_model --key_point_model path/to/key_point_model --base_path path/to/results --input path/to/test_image_folder/images --debug --evalFor the input you can either choose an entire folder of images or a single image. Both times the result will be saved to a new run folder created in the base_path folder. For each image in the input folder a separate folder will be created.
In each such folder the reading is stored inside the result.json file. If there is no such reading, one of the pipeline stages failed before a reading could be computed. Best check the log file which is saved inside the run folder, to see where the error came up. There will also be a error.json file saved to the image folder, which computes some metrics to check without any labels how good our estimate is.
Additionally if the debug flag is set then the plots of all pipeline stages will be added to this folder. If the eval flag is set then there will also be a result_full.json file created. This file contains the data of the individual stages of the pipeline, which is used when evaluating in the script full_evaluation.py.
I prepared two scripts to automatically run the pipeline and evaluations on multiple folders with one command. This allows us to easily conduct experiments for images that we group by their characteristics in different folders.
If they want to be used, make sure to modify the paths inside the scripts, to match with your data.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent