cse515p2

Phase 3

Note: Please run phase 1 task 2 first to cretea descripter database before running any of the following tasks.

Task 1

Arguments:

-p PATH: The path of labeled images.

-unp PATH: The path of unlabeled images.

-k K: The top k of latent semantics.

-t TABLE: The name of table used to store image features.

-meta PATH: The path of metadata for labeled images.

-test(optional) Test mode: The program will print the accuracy, the meta has to be the complete one.

python p3task1.py -p ~/hw/cse515_data/phase3_sample_data/Labelled/Set1 -unp ~/hw/cse515_data/phase3_sample_data/Unlabelled/Set\ 1 -k 30 -t task12 -meta ~/hw/cse515_data/phase3_sample_data/labelled_set1.csv

Task 2:

Arguments:

-p PATH: The path of labeled images.

-unp PATH: The path of unlabeled images.

-k K: The top k of latent semantics.

-c CLUSTER: The number of clusters.

-t TABLE: The name of table used to store image features.

-meta PATH: The path of metadata for labeled images.

-test(optional) Test mode: The program will print the accuracy, the meta has to be the complete one.

-k(optional) # of latent semantics: The default value is 10. It is usually set in test mode to show the effect of k.

python p3task2.py -p ~/hw/cse515_data/phase3_sample_data/Labelled/Set1/ -unp ~/hw/cse515_data/phase3_sample_data/Unlabelled/Set\ 1/ -k 50 -c 5 -t task12 -meta ~/hw/cse515_data/phase3_sample_data/labelled_set1.csv

Task 3:

Arguments:

-k K: The number of outgoing edges.

-lk K: Most K dominant images.

-t TABLE: The name of table used to store image features.

-i PATH: The path of input folder.

-lst ID1,ID2,ID3: The ID of 3 query images.

python p3task3.py -k 10 -lk 5 -t set2 -i ~/hw/cse515_data/phase3_sample_data/Labelled/Set2/ -lst Hand_0008333.jpg,Hand_0006183.jpg,Hand_0000074.jpg

Task 4:

Some arguments for task4:

-c CLASSIFIER: The classifier will be used.

• svm(Support Vector Machine)
• dtree(Decision Tree)
• ppr(Personalized Page Rank)

-m MODEL(optional):

• cm(color moment)
• lbp(local binary pattern)
• hog(histograms of oriented gradients)
• sift(scale-invariant feature transform)

-t TABLE(optional): The name that has been used when creating descriptor database.

-ut TABLE(optional): the unlabeled table.

-k K(optional) the number of latent semantics.

-d METHOD(optional): The method will be used to reduce dimensions.

• svd(Singular value decomposition)
• pca(Principal component analysis)
• lda(Latent Dirichlet allocation)
• nmf(Non-negative matrix factorization)

-l PATH(optional): labeled image folder path

-u PATH(optional): unlabeled image folder path

-meta PATH(.csv)(optional): the path of metadata for labeled images.

-tmeta PATH(.csv)(optional): for unlabeled / test images folder. This is optional. If we can provide test label metadata, this task will show the accuracy.

-limg PATH(optional): Labled raw image path if you do not want to use feature extraction.

-uimg PATH(optional): Unlabeled raw image path if you do not want to use feature extraction.

--svm_pretrained(optional): SVM will save its weight to svm/ folder.

--svm_pretrained PATH(optional): SVM will load the weight and will NOT adjust its weight later.

Example:

• Load image features and labels for labeled data / unlabeled data.
• Use PCA as dimension reduction with k = 20.

python p3task4.py -c svm -m hog -t set1 -d pca -k 20 -ut tSet1 -meta ~/hw/cse515_data/phase3_sample_data/labelled_set1.csv -tmeta ~/hw/cse515_data/phase3_sample_data/unlabelled_set1.csv

Example:

• Load raw image file and labels for labeld data / unlabeled data.
• Use SVD as dimension reduction with k = 100

 python p3task4.py -c svm -meta ~/hw/cse515_data/phase3_sample_data/labelled_set1.csv -limg ~/hw/cse515_data/phase3_sample_data/Labelled/Set1 -uimg ~/hw/cse515_data/phase3_sample_data/Unlabelled/Set\ 1/ -tmeta ~/hw/cse515_data/phase3_sample_data/Unlabelled/unlablled_set1.csv -d svd -k 100

Task 5:

-l LAYERS: The number of layers.

-k HASHES: The number of hashes per layer.

-i ID: The ID of query image.

-t T: The number of most similar images.

-tb TABLE: The name of table used to store image features.

-d METHOD: The method will be used to reduce dimensions.

-dir PATH: The path of images.

--visualize_vector(optional): Visualize the feature vectors.

python p3task5.py -l 5 -k 10 -t 20 -i Hand_0000674 -tb 11k -d hog -dir ~/hw/cse515_data/Hands/

Task 6:

Please run p3task5.py first before p3task6.py.

-c CLASSIFIER: The classifier will be used.

-m MODEL: The model of image feature will be used.

-t TABLE: The name of table used to store image features.

-d METHOD: The method will be used to reduce dimension.

-k K: The top k latent semantics.

In task6, the there are prompts as user to input r(relevant), i(irrelevant), or ?(do not know).

python p3task6.py -c svm -m cavg -d pca -k 20 -t 11k

Phase 1

Task 2:

Give input filepath with argument -i, use color moments with argument -m, and store to table "test".

python p1task2.py -i ~/hw/cse515_data/CSE\ 515\ Fall19\ -\ Smaller\ Dataset -m cm -t test