Knowledge Guided Machine Learning and Computer Vision

Knowledge-Guided Object Detection via Bayesian Networks and Knowledge Graphs (KGB- NCNet) 

This paper has three main contributions:

1. We propose a new approach that combines knowledge graphs and BN to assist with object detection, that integrates knowledge graphs with a BN.

2. We present a technique for creating task-specific knowledge graphs that accurately represent contextual relationships that are essential to object detection ob- ject detection performance.

3. We validate the efficacy of our method by conducting thorough assessments on widely accepted and difficult item detection benchmarks COCO dataset [14]. Our proposed model demonstrates a significant mAP boost, up to 2.02%, relative to the state-of-the-art baseline.

KGGCN: Knowledge-Guided Graph Convolutional Network for Multi-Label Image Classification  

🤖Main Contribution:

➡️ 1.We propose KGGCN, a hybrid model that combines Darknet53 and a knowledge-guided GCN to jointly learn visual and semantic features for better label correlation modelling.

➡️2.KGGCN uses ConceptNet5 and ConceptNet Numberbatch embeddings to build a semantically rich label graph and capture label relationships more effectively.

➡️3.Extensive experiments on VOC 2007 and COCO show that KGGCN achieves state-of-the-art performance and outperforms prior methods in accuracy and robustness. 

KGGCN: Knowledge-Guided Graph Convolutional Network for Multi-Label Image Classification