Relevance Feedback using SVM Active Learning in Python

A Python implementation of SVM-Active-based Relevance Feedback based on paper Image Retrieval with Relevance Feedback using SVM Active Learning

Demo

As the demo below, the accuracy (mAP) of retrieval results is improved after each iteration of relevance feedback:

Requirements

1. Image Dataset

• I use Corel dataset that is very popular in CBIR to demo. The data folder is compressed at Corel.zip in my repository.
• Paths of images are saved as PKL file at paths.pkl.\

2. Features Database

• All Corel images are extracted to 4096d-vectors features using CNN. I use VGG19 as a feature extractor. The features database is compressed at features.zip in my repository.

Main objectives

• As authors: "In content-based image retrieval, relevant feedback is studied extensively to narrow the gap between low-level image feature and high-level semantic concept. In general, relevance feedback aims to improve the retrieval performance by learning with user’s judgements on the retrieval results."
• Active-SVM method that is proposed in the paper not only provides an effective selection of data samples for training SVM but also creates a new ranking function in order to improve retrieval results after each iteration of feedback.

Running

Argparse:

• query_image_path: path of image query to search
• rf: use relevance feedback or not
• rf_loop: number of iterations of relevance feedback
• k_future: number of samples being labeled in future by active learning\

1. Image Retrieval Without Relevance Feedback

python rf.py --query_image_path db/Corel/wl_tiger/108009.jpg

2. Image Retrieval With Relevance Feedback

python rf.py --query_image_path db/Corel/wl_tiger/108009.jpg --rf True --rf_loop 3 --k_future 500