Clever Geek Handbook

How do I apply word2vec to images?

I am learning word2vec model from google. I was able to generate vectors for the corpus text word for a maximum of 300 dimensions. This is a very impressive tool, and the accuracy goes much further with big data.

I'm curious if there is a way to use word2vec to generate vectors in grayscale images. I'm pretty sure the approach is the same, you generate vectors based on pixel intensities and then calculate cosine similarity.

I am trying to create a model to calculate similarity distance in grayscale images. Any library capable of doing this besides word2vec or glove that works in text?

+3


source to share


2 answers


Word2vec is not a good model for images, however I think you really need a dictionary model bag . In the basic image comparison method, you convert the images to a list of key features (like SIFT, SURF, or so on), Then you match clusters of points to each other (like Flann ).



The high number of image functions and the ambiguity of each point representation makes it difficult to use basic single-layer network learning like word2vec for image recognition. You can find the best examples in these tutorials :

+1


source


I agree with you that word2vec is a very impressive tool, but this model learns to predict the next word in some article or news. Overall, I think using word2vec in an image doesn't make sense.



You can use skimage to perform some measure of the image. for example skimage size

+2


source






More articles:


All Articles