I am implementing a MultiLayer Perceptron , I am extracting the features of images using SIFT algorithm of image processing and I pass those features to the neural network , the features of images that I am considering are descriptors ,every image has different length of descriptors , some image has 200 descriptors and some image has 240 descriptors , means it's varying . But neural networks accepts fixed size of input data . How can I pass this type of input to it if it accept varied input then how ?
Asked
Active
Viewed 217 times
0
-
the dimension of the input data needs to be fixed, but you can play around with other parameters – learner May 14 '20 at 06:15
2 Answers
1
I think this is the questions (and answers) you are looking for are below:
- https://ai.stackexchange.com/questions/2008/how-can-neural-networks-deal-with-varying-input-sizes
- How are neural networks used when the number of inputs could be variable?
But in your case, at the end you could get rid of the problem by simply using some image embedding. You could take some neural net that is pre-trained for example, Inception v3 that extracts N features from image, and you will always have constant input, of N features.
Igor
- 43
- 1
- 8
0
I think you could try to pad your data to a fixed size(like the maximum length of your descriptors) vector with all 0 or mean value of this vector.
邓剑波
- 11
- 1