I have seen the scale of at least 1000 s for cnn. I know it depends on many factors like the image and its details but is there roughly any estimate that can determine the number of samples is required to apply CNN reliably?
Have you seen 100 of images applied for CNN?
For your current data I suggest simple SVM is enough. However., for the case of CNN via DL you may have a look on large datasets as here : https://pathmind.com/wiki/open-datasets
Dr Sharma thank you for suggestion. By the way, what you mean by fine tuning? and pre-trained model?
Muhammad Ali I need to classify images and SVM is not straightforward in this case.
By the way what you mean by "DL"?
Dear Zeinab
100 number of images is quite low for a CNN algorithm. Appropriate number of samples depends on the specific problem, and it should be tested for each case individually. But a rough rule of thumb is to train a CNN algorithm with a data set larger than 5,000 samples for effective generalization of the problem. To increase your samples you can use data augmentation. This method aims at artificially expanding the size of a training data set by creating modified versions of models (randomly rotating the image, adding reflections, zooming in, adding a colour filter, etc). Data augmentation is a commonly-used approach in image processing applications to create variations of the images that can improve the ability of the trained network for generalizations to new images.
- Badges
- Science topic