HW2 of Cognitive Computing, 2019 fall, NTU CSIE
In this homework, we experiment with images from 30 categories, with 20 images in each category, our goal is to design/extract features from the image and retreive images using the features. We experiment with color, texture/shape and local feature in this homework.
We use MAP as our evaluation metric. We do it in leave-one-out fashion and report over MAP and MAP for each category.
We experiment with global/local color histogram, global/local color moment and color auto-correlogram, the results are reported in hw2.ipynb
We experiment with Gabor feature, local binary pattern, histogram of oriented gradient, and shape index, the results are reported in hw2.ipynb
We experiment with SIFT desciptors, the result are reported in hw2.pynb
These 2 files contain the functions for extracting feature and functions for running experiments
This notebook would generate the feature database for exeriments, it takes pretty long to generate features, mostly spent on generating auto-correlograms, you could comment auto-correlogram out to speed up or email me to ask for the final FeatureDatabase.pkl
All the experiments and results are all reported in hw2.ipynb
,long with some comments and findings of the experiments.
- Python 3.7.3 :: Anaconda custom (64-bit)
- SaKaTetsu - Initial work - SaKaTetsu