Enhancing Referring Expression Segmentation Through Positional Context
Tuba Amreen Darwesh
, Anita Harsoor
Index Terms: GRES ,NLP ,Ref Exp Segmentation
Abstract-Referring Expression Segmentation (RES) is a single procedure that involves segmentation construction. mask for an item recognised by a certain linguistic communication. This approach focusses mostly on expressions with a single target, and single expression fits the intended object. Conversely, (GRES) increases the RES task's scope by allowing phrases that indicate any number of targets items. This covers situations with several targets—one target, none at all, and targets. GRES aims to overcome the limitations that RES datasets have, and approaches, hence improving its relevance in a variety of settings.(RES) as well as (GRES) enable improved understanding of visuals by producing human language-based segmentation masks an explanation Using databases such as gRefCOCO and techniques such as ReLA, which bridge the gap between language comprehension as well as computer vision.. The primary objective of RES and GRES is to generate segmentation masks for objects referenced in natural language expressions within images.The aims to enhance image understanding by accurately delineating objects based on linguistic descriptions, thereby improving object recognition and scene interpretation. The implementation of (RES) and (GRES) involves dataset preparation with paired images and language expressions. Neural network models like CNNs or transformer-based architectures are trained on these datasets to learn the relationship between image regions and language descriptions. Trained models generate segmentation masks for referenced objects in new images. Evaluation assesses segmentation accuracy and informs iterative refinement of model architecture. Finally, the optimized model is deployed for applications like as image understanding and human- computer interaction.
"Enhancing Referring Expression Segmentation Through Positional Context", IJSDR - International Journal of Scientific Development and Research (www.IJSDR.org), ISSN:2455-2631, Vol.9, Issue 9, page no.276 - 287, September-2024, Available :https://ijsdr.org/papers/IJSDR2409030.pdf
Volume 9
Issue 9,
September-2024
Pages : 276 - 287
Paper Reg. ID: IJSDR_212485
Published Paper Id: IJSDR2409030
Downloads: 000347088
Research Area: Computer Engineering
Country: Kalaburagi, Karnataka, India
ISSN: 2455-2631 | IMPACT FACTOR: 9.15 Calculated By Google Scholar | ESTD YEAR: 2016
An International Scholarly Open Access Journal, Peer-Reviewed, Refereed Journal Impact Factor 9.15 Calculate by Google Scholar and Semantic Scholar | AI-Powered Research Tool, Multidisciplinary, Monthly, Multilanguage Journal Indexing in All Major Database & Metadata, Citation Generator
Publisher: IJSDR(IJ Publication) Janvi Wave
