How to make basic image classification python script without training

What is the name of the process (in python) who classify image (from webcam in my case) based on pre train model but without time cost trainning with my custom image ?

My question start with this code a “simple” knn features matcher WITHOUT TRAINING PART, based on ML5js example from coding train. You catch some images from your webcam, and directly the script labeled and included them in your model which start reacting from the first image recorded.

In web editor with the script : p5.js Web Editor

I want to make this in python in order to comminucate in a second time with unity game in C#. Basicly i want to make a gesture control game but the teaching part have to be part of the game but not taking more than a minute or two. (All include in a physical ML teaching totem for audience…)

What i have found so far in Python :

To be efficient, a model have to be trainned => For me i don't want it to be efficient for general case but only with something like 80 images for 4 differentes classes (Left, right, up, down).

When i am looking about features extractor, i always find a comparaison between two single images. => I want to compare the webcam stream with a bunch of images (About 80img in 4 classes)

I have found the term “bag of feature”, but it does not give me anything on web.

I am only looking for a documentation near form what i am looking or maybe an explanation.

that would be “inference”, which is the forward pass through a neural network.

backward passes through a neural network are called “training”.

you seem to want a pre-trained DNN for object detection.

1 Like

Thanks for your quick reply.
Inference is one of the word i extract from this js code :slight_smile: But i don’t find anything close to this method (Custom image dataset with pre-trained model).
Every thing i find have to be trained and take more time than i have for my final goal.
Could you help me more by indicate me a documentation, or example maybe ? :slight_smile:

After some research, i don’t find anything about image recognition just to compare the stream from the webcam and somes images in my data set.

I find something to compare to single image with features for example.

Each time i am looking for image recognition, it gives me face or pose recognition but never to compare the all image. Do you have any idea of word that describe this process ?


sounds like “comparing shapes”

Yes off course, but it missed a good research word to find something relevant. It appears that “Camera classifier” is better than every thing like “Webcam forward propagation” or even “webcam classifier”.
I’ve found this that it appears to be more that i am looking for : GitHub - NeuralNine/camera-classifier: A Python application that uses camera input to train a SVM to respond to specific actions.

and, – do you understand if it will be helpful or not ?

sorry to say so, but that’s utter “cargo cult bs”.
(you seem to have no idea, what you’re talking about)

want my 2 ct. ?
let’s close this. you go and take some basic cv courses, read books, papers, learn to code and try out (and evaluate) ideas.


It appears that you have forgotten the meanning of being respectfull. Yes i am learning, not sorry for that.