Warning: this project is based on the assumption that you use projects Search Activities and Image retrieval to obtain the images to be analyzed.
Requirements
- [Node.js]
- [NPM]
- [All necessary to run TensorFlowJS (see here)]
How to install the project?
Just run npm install
How to run the project?
If you want to run only for a particular configuration given in file generalConfig.json, use the command npm start
If you want to run only for multiple configurations given in file experimentationsConfig.json, use the command npm experimentation
Where can I change my parameters?
In the root folder, in the files generalConfig.json for using a particular configuration, experimentationsConfig.json for using multiple configurations and in modelsConfig.json to configure object detection/classification model parameters
Where can I find the results?
In the folder resultFiles. More precisely, in the folders named using your configuration (ex: inception_resnet_v2 google 50 sum)
The format of result files is JSON and is the following:
{
"query": "answer_the_phone",
"data": [
{
"label": "cellular telephone", //The object detected using a particular model
"relevance": 180.73915372416377, //The aggregated score computed using multiple images
"correct": null //attibute only used to compute performance metrics (see groundTruthModel folder and performanceMetrics.js)
},
...
]
}
Where are the files responsible for computing performances?
In the root folder, in the file performanceMetrics.js