We've been using the ORL dataset for a while now, but it's a pretty easy dataset and we need to branch out anyway. Some datasets that I know of:
FERET
MNIST
Yale
VIVA
Some datasets are very large and it may be impractical to download them to your computer, and some datasets may only be accessible by request. Just use what you can; we'll be able to access and store whatever datasets we need through the lab.
For each dataset that you find, you'll need to make sure that the directory structure is compatible with our scripts. You'll either have to rearrange the dataset or we can try to adapt our scripts if it isn't too difficult. Once you can run the scripts, you need to test each algorithm (PCA, LDA, ICA) with different amounts of images removed from the training set (1 per class, 3 per class, 5 per class, etc.), and you need to record the accuracy (%) and run-time (s) of each test case.
We've been using the ORL dataset for a while now, but it's a pretty easy dataset and we need to branch out anyway. Some datasets that I know of:
Some datasets are very large and it may be impractical to download them to your computer, and some datasets may only be accessible by request. Just use what you can; we'll be able to access and store whatever datasets we need through the lab.
For each dataset that you find, you'll need to make sure that the directory structure is compatible with our scripts. You'll either have to rearrange the dataset or we can try to adapt our scripts if it isn't too difficult. Once you can run the scripts, you need to test each algorithm (PCA, LDA, ICA) with different amounts of images removed from the training set (1 per class, 3 per class, 5 per class, etc.), and you need to record the accuracy (%) and run-time (s) of each test case.
Further reading: http://face-rec.org/databases/