kirle / visualsearch_flowerfamilyidentification

Exploring a visual search solution for flower family identification using pretrained CNNs and embedding learning techniques.

VisualSearch_FlowerFamilyIdentification

Using a dataset adapted from the Oxford Flowers Dataset, 1262 images from 24 categories were analyzed. Two tasks were developed. The first task involved computing embeddings, measuring similarity, and ranking images. The second one focused on embedding learning using contrastive or triplet loss and diverse data sampling and compare the performance metrics. Also, a method for comparing two images and decide if their belong to the same class was developed using this ideas. Results demonstrate effective visual search for flower identification using pretrained CNNs and embedding learning.