<![CDATA[Abstract—Image captioning aims to automatically generate textual descriptions that accurately represent visual content. This study proposes a Sequence-to-Sequence image captioning framework that integrates an InceptionV3 encoder with a Vanilla Transformer decoder. The InceptionV3 network is employed to extract visual features from images, while the Vanilla Transformer generates captions by modeling contextual relationships between visual and textual representations. Experiments were conducted on the Flickr30k dataset using two data split scenarios (80:10:10 and 70:15:15) and two training configurations to evaluate the effectiveness of the proposed framework. Model performance was assessed using BLEU, METEOR, ROUGE-L, and CIDEr metrics. The best results were achieved using the 80:10:10 train-validation-test split and training configuration 1, obtaining BLEU-1, BLEU-2, BLEU-3, and BLEU-4 scores of 0.4252, 0.2946, 0.2176, and 0.1601, respectively, along with METEOR, ROUGE-L, and CIDEr scores of 0.4155, 0.4681, and 0.5059. These findings demonstrate that the proposed InceptionV3–Vanilla Transformer architecture is effective in generating accurate and contextually relevant image captions on the Flickr30k dataset. Keywords—Image Captioning, InceptionV3, Vanilla Transformer, Sequence-to-Sequence, Flickr30k.]]>
Copyrights © 2026
