A Node-RED node that calculates the cosine similarity, between a set of stored vectors and provided vector values.
This node is a significantly modified version of 🔗 @good-i-deer/node-red-contrib-cosine-similarity and doesn't maintain compatibility with that node.
✔ input data - accepts payload direct from 🔗 @smcgann/node-red-face-vectorization-plus.
✔ Supports multiple stored vector locations: file, flow context, or global context.
✔ Stored vectors now also include names and image file data, for meaningful results.
✔ Returns sorted results, filtered by a configurable similarity threshold.
✔ Generates an error message if no results meet threshold.
✔ More robust error handling with validation for missing and invalid data structures.
✔ Supports runtime configuration through msg.cosineOptions for dynamic settings.
✔ Includes metadata in msg.cosineConfig (threshold used, file type, file path).
✔ Easier integration into Node-RED flows.
This node calculates cosine similarity between two sets of vectors. One passed in by msg.payload and one stored in a file or context variable.
It returns the results in msg.payloadas an array of objects, where the similarity is above set threshold.
input = array[2] // (2 faces)
0: array[0.12, 0.34, 0.56, "..."] // 512 vectors for each face.
1: array[0.11, 0.22, 0.33, "..."]
// Stored as a nested object, with each key representing an individual person.
storedVectors =
{
"Adam": {
"/full-path/Adam/Adam-01.jpg": [0.12, 0.34, 0.56, "..."], // 512 vectors for each face.
"/full-path/Adam/Adam-03.jpg": [0.23, 0.45, 0.67, "..."],
"/full-path/Adam/Adam-02.jpg": [0.34, 0.56, 0.78, "..."]
},
"Alison": {
"/full-path/Alison/Alison-01.jpg": [0.11, 0.22, 0.33, "..."],
"/full-path/Alison/Alison-02.jpg": [0.44, 0.55, 0.66, "..."]
}
// repeated for x number of people
}
output = array[2] // (2 matches above threshold)
[
{ "Alison": { "/full-path/Alison/8.jpg": 0.7265643591861766 } },
{ "Peter": { "/full-path/people/Peter/9.jpg": 0.6443388973714721 } }
]Either use the Edit Menu - Manage Palette option to install, or run the following command in your Node-RED user directory - typically ~/.node-red
cd ~/.node-red
npm install @smcgann/node-red-cosine-similarity-plusRestart your Node-RED instance
- The input is an array of vector arrays. (created by @smcgann/node-red-face-vectorization-plus)
- Allows overriding node config settings dynamically.
- Example:
{
"threshold": 0.4,
"fileType":"path",
"file":"/home/pi/vectortest.txt"
} 
- The name of the node displayed on the screen.
- Results with a value equal to or above the specified threshold will be returned. Set the threshold to 0 to include all matches.
- File or context path of file or variable, that will be compared with the input vector array. Cannot be empty.
- The output is an array of name objects. Each containing the path to the original face image file and a percentage similarity to the input vectors. Where no match is above the set threshold, an empty object is created to maintain the index order between names and bounding boxes.
- The configuration used for the calculations.
S.McGann → Modified Version.
GOOD-I-DEER in SSAFY(Samsung Software Academy for Youth) 9th
S.McGann → Modified Version
Copyright Samsung Automation Studio Team under the Apache 2.0 license