Color-Attributes-Related Image Retrieval Student: Kylie Gorman Mentor: Yang Zhang
Fix Code Initially calculating PCA and GMM independently Calculate GMM based on PCA results
Steps
Improved Steps
Compare New and Previous Results Improved Results Compare New and Previous Results
HSV Results Original Average: ~25% New Average: ~45%
RGB Results Original Average: ~20% New Average: ~50%
CIELAB Results Original Average: ~17% New Average: ~42%
Combined Results Average: ~9% Early Fusion Did Not Work Possibly Requires Debugging
New sets to include separate attributes such as object recognition New Data Sets New sets to include separate attributes such as object recognition
New Data Birds 200 Flowers 102 Cartoon 200 species/categories with 6,033 images total Flowers 102 102 categories with 40-258 images per category 8189 images total Cartoon 590 images total
Flowers 102 Part One Part Two Part Three Get Feature Matrices with Color Moments Calculate PCA and GMM of training data: 1,020 images Part Two Get Feature Matrices with Dense SIFT Calculate PCA and GMM of training data: 100 images Part Three Use new Color Descriptor
Future Goals Compare our color moment plus Dense SIFT against new color descriptor and Dense SIFT If no improvement, determine why Follow same steps with Bird 200 data set