University of Twente Student Theses

Login

LCNet: Learnable Label Correlation Network for Multi-Label Image Classification

Scholten, M.H.D. (2024) LCNet: Learnable Label Correlation Network for Multi-Label Image Classification.

Full text not available from this repository.

Full Text Status:Access to this publication is restricted
Embargo date:1 December 2025
Abstract:In computer vision, multi-label image classification is a well known topic which can be applied in many real-world applications. In classical multi-label image classification, only the spatial features of images are used as input of classification models. In this paper we introduce LCNet: A MultiLabel Classification model, which includes a learnable semantic graph followed by a novel decoder to fuse the multiple modalities. Our decoder is stack-able which allows the model to understand deeper relations between the modalities. In addition, we design a feature to reduce spatial resolution loss called Crop-Forwarding. Crop-Forwarding allows higher resolution images to be forwarded without using higher resolution spatial feature extractors. Extensive experiments conducted on several multi-label classification benchmarks, Pascal-VOC and MS-COCO, demonstrate that our solution significantly improved the state-of-the-art results. Our proposed method achieves mAP 91.5% on MSCOCO and 97.7% on Pascal-VOC. Especially, our model outperforms the state-of-the-art models on a small datasets such as the synthetic-fiber rope damage dataset, resulting in a new top score of 88.2%.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:50 technical science in general, 54 computer science
Programme:Embedded Systems MSc (60331)
Link to this item:https://purl.utwente.nl/essays/104715
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page