IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 13, No. 1, March 2024, pp. 23~34

ISSN: 2252-8938, DOI: 10.11591/ijai.v13.i1.pp23-34  23

A study on attention-based deep learning architecture model for

image captioning

Dhomas Hatta Fudholi, Umar Abdul Aziz Al-Faruq, Royan Abida N. Nayoan, Annisa Zahra
Department of Informatics, Universitas Islam Indonesia, Yogyakarta, Indonesia

Article Info ABSTRACT

Article history: Image captioning has been widely studied due to its ability in a visual scene
understanding. Automatic visual scene understanding is useful for remote
Received Nov 20, 2022 monitoring system and visually impaired people. Attention-based models,
Revised Mar 17, 2023 including transformer, are the current state-of-the-art architectures used in
Accepted Mar 27, 2023 developing image captioning model. This study examines the works in the
development of image captioning model, especially models that are developed
based on attention mechanism. The architecture, the dataset, and the
Keywords: evaluation metrics analysis are done to the collected works. A general flow of
image captioning model development is also presented. The literature search
Attention process carried out on Google Scholar. There are 36 literatures used in this
Encoder-decoder study, including a specific image captioning development in Indonesian. It is
Evaluation metrics done to take one point of view of image captioning development in a low
Image captioning resource language. Studies using transformer model generally achieves higher
Transformer evaluation metric scores. In our finding, the highest evaluation scores on the
consensus-based image description evaluation (CIDEr) c5 and c40 metrics are
138.5 and 140.5 respectively. This study gives a baseline on future
development of image captioning model and brings the general concept of the
image captioning development process including a picture of the development
in low resource language.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Dhomas Hatta Fudholi
Department of Informatics, Universitas Islam Indonesia
Kaliurang St. Km. 14.5 Yogyakarta 55584, Indonesia
Email: [email protected]

1. INTRODUCTION
Image captioning is the ability to describe the contents of an image in the form of sentences [1]. This
ability requires methods from two fields of artificial intelligence, namely computer vision to understand the
content of a given image and natural language processing (NLP) to convert the content in the image into
sentence form. Due to the advancement of deep learning models in these two fields, image captioning has
received a lot of attention in recent years. On the computer vision side, the improvement of the convolution
neural network (CNN) architecture and object detection contributed to the improvement of the image
captioning system. On the NLP side, more advanced sequential models, such as attention-based recurrent
networks, result in more accurate text generation.
Most successful image captioning uses an encoder-decoder approach inspired by the sequence-to-
sequence model for machine translation. This framework uses CNN + recurrent neural network (RNN), where
CNN is used as an image encoder that extracts region-based visual features from the input image and RNN is
used as a caption decoder to make sentences [2]. With the development of machine translation, a new
architecture emerged, namely attention, which has become a state-of-the-art method in the NLP field and

Journal homepage: http://ijai.iaescore.com

24  ISSN: 2252-8938

makes image captioning models. Transformer [3] is an attention-based architecture that largely being adopted
in the field.
This study aims to discover state-of-the-art method in image captioning, especially attention-based
architecture, including transformer. We also take Indonesian as one of the low resource languages as a case
study to give a picture of how mature the development of image captioning model in such language. An analysis
of the model and architecture used, as well as the results of the study was carried out. Additionally, we develop
a general concept that indicate what needs to be done, what techniques and data are typically used in image
captioning research. The results of this study are expected to be a reference for the image captioning method
in future research and provide recommendations for the best method for doing image captioning.

2. METHOD
2.1. Research flow
The systematic literature review presented in this paper has two main steps in its methodology. The
first step is to formulate and conduct a search for related literatures. The second step is to formulate, conduct,
and discuss the literature analysis from several points of view. Details of the methodological steps carried out
will be explained in the next section.

2.2. Searching method

The searching process for related literatures was carried out in three parts. The process uses the help of
the Google Scholar search engine. The searching process will select literatures that meets the following criterias:
i) the scientific publication literature uses the attention mechanism model as the basis for doing image captioning;
and ii) the literature provides details on the implementation of the method in image captioning task.
The first part of the searching process uses three main keywords: image captioning, transformer
models, and attention mechanism. The scientific publication taken in this part is publications within the time
frame between 2017 and 2020. This time frame is taken to depicts the growth and the development of the
attention-based mechanism in image captioning task. The second part of the searching process is carried out
with the same keywords as the first part, but the time frame is between 2021 and 2022. In this part, only top
four cited open access publication is taken. The aim of the second part is to get some knowledge how attention-
based model in image captioning still can be improved. The last part is slightly different from the previous
parts. The third part focuses on the study of image captioning in Indonesian. Indonesian image captioning is
used as the keyword. This part intends to search studies using Indonesian language datasets from 2017 to 2022
that use attention-based model including transformer.

2.3. Analytical method

Selected literatures are analyzed from several points of view. There are five analyzes to be carried
out, namely architectural analysis, method analysis, dataset analysis, metric evaluation analysis, and general
modeling model analysis of image captioning. The analysis was carried out with the aim of obtaining
information that could support the growth of research in this study. The results of the analysis are presented in
tabular form with a detailed explanation of the analysis.

3. RESULTS AND DISCUSSION

Using the keywords given in the literature search, 27 literatures were obtained in the first step. Of the
27 literatures, 25 literatures met the search criteria, and two literatures did not meet the search criteria. One
literature does not meet the specified time range between 2017 to 2020 and one last literature does not use the
attention mechanism model. In the second search process, we selected 4 literatures from 2021, and 4 others
from 2022 that are on the first five pages of search results and can be accessed. While our third search resulted
in 4 literatures that match our criteria.
Table 1 (see in Appendix) shows the comparison between the selected literatures according to the
search criteria. The table provides a comparison between the architecture, the methods, and the evaluation
metrics results. The evaluation metrics presented are bilingual evaluation understudy (BLEU), metric for
evaluation for translation with explicit ordering (METEOR), recall-oriented understudy for gisting evaluation-
longest common subsequence (ROUGE-L), consensus-based image description evaluation (CIDEr), and
semantic propositional image caption evaluation (SPICE). The higher the score, the more accurate the predicted
caption based on the referenced caption. The elaboration of the architecture, the dataset, and the evaluation
metrics analysis are presented in the following section. The literature comparison for papers that used
Indonesian dataset is presented in Table 2.

Int J Artif Intell, Vol. 13, No. 1, March 2024: 23-34

Int J Artif Intell ISSN: 2252-8938  25

Table 2. Literature review result for Indonesian datasets

Literature Architecture Method Evaluation Score
[34] Attention ResNet101 and LSTM BLEU-1,2,3,4: 67.8, 51.2, 37.5, 27.4; CIDEr: 99.0
[35] Attention EfficientNet Average of the BLEU train score: 73.39; Average of the
BLEU validation score: 24.51
[36] Transformer Pretrained ResNet BLEU-1,2,3,4: 56.00, 41.17, 29.42, 20.57; METEOR: 19.50;
ROUGE-L: 44.16; CIDEr: 57.26
[37] Transformer Multi-head attention and CNN Average BLEU-1,2,3,4: 78.05, 68.21, 61.89, 52.09

3.1. Architecture analysis

In this section, the elaboration of different approaches in developing image captioning model are
presented. The analysis of the elaboration is grouped into two. The first one is the vanilla attention mechanism,
and the second one is the transformer which uses a more specific attention mechanism called self-attention.

3.1.1. Attention mechanism

Of the image captioning methods that use the attention mechanism, many use a CNN encoder and an
RNN or long short-term memory (LSTM) decoder. Among them are the hierarchical attention network (HAN)
[7] which pays attention to semantic features of various levels that make it easy to predict different words based
on different features, while the multivariate residual module (MRM) makes it easy to extract relevant relations
from various features. There are also other methods that also use the same encoder and decoder, namely,
attention on attention (AoA) [8], scene graph auto-encoder (SGAE) [12], Adaptive attention through visual
sentinel [20], policy optimization gradient SPIDEr [22], and recurrent fusion network (RFNet) [15]. In
addition, research that utilizes spatial and semantic information using attention mechanisms, namely the graph
convolutional network-long short-term memory (GCN-LSTM) [17], and spatial and channel wise attentions in
a CNN (SCA-CNN) [21] which can score high when compared to state-of-the-art models. LSTM-P [5] uses an
RNN-based language model that presents a novelty, namely the exploitation of the pointer mechanism to
accommodate dynamic word generation through an RNN-based language model and word copying from the
object being studied. hierarchy parsing (HIP) [6] which integrates a hierarchical structure into an image
encoder. HIP functions as a feature refiner producing a rich and multi-level representation of the image.
Unsupervised image captioning [16] uses an encoder, generator, and descriptor. In this method, CNN will
encode the input image into a feature representation, then the LSTM as a generator will decoding the image
representation into a sentence that describes the image content, while the LSTM descriminator is tasked with
distinguishing the original sentence from the sentence generated by the model. The combination of top-down
and bottom-up attention mechanisms [19] where bottom-up based on Faster R-CNN processes the image area
and converts it into feature vectors, while top-down determines feature weighting. Research [23] overcomes
the variation and ambiguity of image descriptions with the convolution technique. Research on updating the
long short-term memory (LSTM-A) architecture [25] has succeeded in integrating attributes in CNNs + RNN
framework image captioning. Having a proven performance in generating meaningful sentences and being very
successful in advancing state-of-the-art, the attention mechanism is still being used in recent research [33].
Prophet attention [33] was introduced for calculating the ideal attention weights towards image region by using
the future information.
According to our search, attention is rarely used in Indonesian image captioning from 2017 as we only
found two papers that matched our criteria. To produce the next word, adaptive attention [34] was used to
determine when and at which part of the image should be focused on by using translated Microsoft Common
Objects in Context (MS COCO) and Flickr30k datasets. Research [35] applied visual attention mechanism to
their model to produce a caption for the image that makes greater sense. As the result, their model was able to
give a sensible and detailed caption in the local tourism domain.

3.1.2. Transformer
Transformer works as an encoder-decoder architecture that uses an attention mechanism. The
captioning transformer (CT) study [18] was developed to overcome the problem of image captioning which is
often developed using an LSTM decoder. Although good at remembering sequentially, LSTM has a
complicated sequential problem in terms of timing. CT only has an attention module without a time
dependency, so this model not only remembers sequence dependencies, but can also be trained in parallel.
Research [1] uses a spatial graph encoding transformer layer with a modified encoding transformer
arrangement and an implicit decoding transformer layer which has a decoder layer and an LSTM layer in it to
overcome the structure of the semantic unit of the image and each word in a different sentence. Research [4]
improves image encoding and text prediction by using meshed transformer with memory to get low- and high-
level features so that it can predict images that are not even in the training data. Multimodal transformer [2] is
composed of an image encoder and a text decoder simultaneously capturing intra and inter-modal interactions
A study on attention-based deep learning architecture model for image captioning (Dhomas Hatta Fudholi)
26  ISSN: 2252-8938

such as the relationship of words, objects, words in attention blocks that can produce captions accurately.
Boosted transformer [9] utilizes semantic concepts (CGA) and visual features (VGA) to enhance the
description of the resulting image. Personality-captions [13] uses TransResNet and a dataset that supports
personality differentiation to produce image descriptions that are closer to humans. Conceptual dataset [14]
was also developed using Inception-ResNetv2 as a feature extractor and transformer to perform image
captioning. Another study with encoder and decoder transformer using object spatial relationship model [10]
was built by explicitly including spatial relationship information between input objects detected from attention
geometric. EnTangled transformer [11] was developed to exploit all semantic and spatial information from
images.
Since various transformer-based models have achieved promising success on the image captioning
task [31], recent research has still widely used it. dual-level collaborative transformer [26] was proposed to
complement region and grid features for image captioning by applying intra-level fusion via comprehensive
relation attention (CRA) and dual-way self attention (DWSA). Global enhanced transformer (GET) [27] makes
it possible to obtain a more comprehensive global representation, which guide the decoder in creating a high-
quality caption. Caption transformer (CPTR) [28], as a full transformer model, is capable of modeling global
context information throughout encoder at every layer. Transformer-based semi-autoregressive model for
image captioning, which keeps the autoregressive property in global and non-autoregressive property in local,
tackles the heavy latency during inference issue that is caused by adopting autoregressive decoders [29]. Spatial
and scale-aware transformer (S2 transformer) [30] explores both low-level and high-level encoded features
simultaneously in a scale-wise reinforcement module and learns pseudo regions by learning clusters in a
Spatial-aware Pseudo-supervised module. Relational transformer (ReFormer) [31] was proposed to improve
the quality of image captions by generating features that have relation information embedded, as well as
explicitly expressing pair-wise relationships between images and their objects. While research [32] used a
transformer-based architecture called attention-reinforced transformer to overcome the problem of cross
entropy limiting diversity in image captioning.
For research with Indonesian dataset, we only found two paper that use transformer in their study,
[36] and [37]. The result of research [37] showed that the implementation of the transformer architecture
significantly exceeded the results of existing Indonesian image captioning research. In addition, the use of
EfficientNet model obtains better results than InceptionV3. Research [36] has different approach, which use
ResNet family as the base of visual feature extraction.

3.2. Dataset analysis

From the analysis conducted on existing studies, five main datasets were generaly used for image
captioning, namely conceptual captions, MS COCO, Flickr8K, Flickr30K, and a specially made dataset for
local tourism domain. The composition of the four datasets and studies that utilize these datasets can be seen
in Table 3. On Table 3, we can see the detail of the number of images in the training, the validation, and the
testing dataset. Furthermore, we can see the number of annotations for each image in different dataset.

Table 3. Dataset composition

Dataset Training Validation Testing Annotations for each Literature
image
Conceptual 3,318,333 15,840 12,559 5 [14]
MS COCO 113,287 5,000 5,000 5 All literature that used English
datasets
Flickr30K 29,783 1,000 1,000 5 [13], [20], [21], [24]
Flickr8K 6,000 1,000 1,000 5 [21]
Specially made (Local Tourism 1,356 340 - 1 [35]
Domain)

3.2.1. Conceptual captions

Conceptual captions [14] were created using the Flume pipeline. This pipeline processes billions of
internet pages in parallel. From these web pages, extraction, filtering, and pairing processes (images,
descriptions) were carried out. The filtering and processing steps are divided into four, namely image-base
filtering, text-based filtering, image & text-based filtering, and text transformation with hypernymization.

3.2.2. MS COCO
MS COCO [38] is a dataset from Microsoft COCO which is a large dataset containing object
detection, segmentation, and captioning. This dataset has 328,000 images with a total of 2,500,000 labels, 80

Int J Artif Intell, Vol. 13, No. 1, March 2024: 23-34

Int J Artif Intell ISSN: 2252-8938  27

object categories, and 91 object categories. This dataset is divided into training, validation, and testing data, as
in Table 3, using Karpathy's splits [39].
The image captions dataset in MS COCO consists of two dataset collections. The first dataset, MS
COCO c5, has five text references for each image in the MS COCO dataset training, validation, and testing.
The second dataset, MS COCO c40, has 40 text references and randomly selects 5,000 images from the MS
COCO testing dataset. MS COCO c40 builds on the many automated evaluation metrics that give results that
achieve higher correlations than human judgments when given more references [40].

3.2.3. Flickr30K and flickr8K

Flickr30K [41] is a popular dataset used as a benchmark for text generation and retrieval. Flickr30k
has 31,783 images focused on humans and animals, as well as a total of 158,915 English subtitles for these
images to reference. While Flickr8K [42] has a total of 8000 images collected from Flickr. Each image in the
dataset has five human-annotated captions.

3.2.4. Indonesian datasets

To translate the English MS COCO dataset into Indonesian, two methods were used: Google Translate
[36], [37] and manual translation [34]. However, the results of the google translate translation are not very
good, so research [34] used both of those two methods in their study to get a good Indonesian dataset. Study
with a specific domain, requires a specially made dataset because it has not been available before. Research
[35] collected a total of 1,696 local tourism-related images from Google search engines.

3.3. Evaluation metrics analysis

From the analyzed literature, five different evaluation metrics are commonly used to evaluate image
captioning: BLEU, METEOR, ROUGE, CIDEr and SPICE. These metrics mainly measure the similarity
between generated and reference captions through word overlap. Especially in SPICE, it uses “scene graph” to
measure the similarity [43]. BLEU, METEOR, and ROUGE-L are evaluation models originally developed to
assess the performance of Machine Translation. In recent years CIDEr and SPICE were developed specifically
to evaluate image captions and showed more success than the previous ones [44].
The evaluation is done to measure the quality of a candidate caption ci given a set of reference captions
𝑆𝑖 = {𝑠𝑖1 , 𝑠𝑖2 , … , 𝑠𝑖𝑗 } ∈ S. When the sentences are represented using sets of n-grams, ωk ∈ Ω is a set of one or
more ordered words. For the candidate sentence ci ∈ C, hk(sij) or hk(ci) denotes the number of times an n-gram
ωk occurs in a sentence sij.
BLEU calculates the n-gram overlap between candidate and reference texts to evaluates candidate
texts. The BLEU score is calculated by the geometric mean of the modified n-gram score accuracy. The score
is multiplied by a short penalty factor to give a "punishment" to short sentences so that the evaluation results
are more representative [44]. The clipped n-gram precision between sentences is computed at the corpus level
as shown in (1) [40]. In this case, k represents the set of possible n-grams of length n. While 𝐶𝑃𝑛 favors short
sentences as it’s a precision score. It is also used a brevity penalty as in (2) to favor a short sentence. Here, 𝑙𝐶
represents the total length of candidate sentences 𝑐𝑖 ’s and 𝑙𝐶 represents the corpus-level effective reference
length. For the brevity penalty, we use the closest reference length whenever multiple references exist for a
candidate sentence. To calculate the overall BLEU score, a weighted geometric mean of the individual n-gram
precision is applied as in (3) where the values of 𝑁 are 1, 2, 3, 4 and 𝑤𝑛 is usually constant for all 𝑛.

∑𝑖 ∑𝑘 min⁡(ℎ𝑘 (𝑐𝑖 ), 𝑚𝑎𝑥

𝑗∈𝑚ℎ𝑘 (𝑠𝑖𝑗 ))
𝐶𝑃𝑛 (C, S) = ∑𝑖 ∑𝑘 ℎ𝑘 (𝑐𝑖 )
(1)

1, 𝑖𝑓⁡𝑙𝐶 > 𝑙𝑆
𝑏(𝐶, 𝑆) = { 1−𝑙𝑆 ∕𝑙𝐶 (2)
𝑒 , 𝑖𝑓⁡𝑙𝐶 ≤ 𝑙𝑆

𝐵𝐿𝐸𝑈𝑁 (𝐶, 𝑆) = 𝑏(𝐶, 𝑆)𝑒𝑥𝑝⁡(∑𝑁

𝑛=1 𝑤𝑛 𝑙𝑜𝑔 ⁡𝐶𝑃𝑛 (𝐶, 𝑆)⁡) (3)

Metric for Evaluation for Translation with Explicit Ordering (METEOR) evaluates the candidate text
based on overlapping unigrams between candidate and reference texts. This corresponds to a unigram based
on meanings, exact and stemmed forms [44]. During calculating the alignment between the words in the
candidate and reference sentences, the number of contiguous and identically ordered chunks of tokens in the
sentence pair (𝑐ℎ) is minimized. This evaluation is conducted using the default parameters 𝛾, 𝛼 and 𝜃. So, based
on a set of alignments (𝑚), the METEOR score is derived from the harmonic mean of precision (𝑃𝑚 ) and recall
(𝑅𝑚 ) between the candidate and reference with the best score [40], see (4)-(8).

A study on attention-based deep learning architecture model for image captioning (Dhomas Hatta Fudholi)
28  ISSN: 2252-8938

𝑐ℎ 𝜃
𝑃𝑒𝑛⁡ = 𝛾⁡ ( ) (4)
𝑚

𝑃𝑚 𝑅𝑚
𝐹𝑚𝑒𝑎𝑛 ⁡ = ⁡ (5)
∝𝑃𝑚 +(1−∝)𝑅𝑚

|𝑚|
𝑃𝑚 ⁡ = ⁡ ∑ (6)
𝑘 ℎ𝑘 (𝑐𝑖 )

|𝑚|
𝑅𝑚 ⁡ = ⁡ ∑ (7)
𝑘 ℎ𝑘 (𝑆𝑖𝑗 )

𝑀𝐸𝑇𝐸𝑂𝑅 = (1 − 𝑃𝑒𝑛)𝐹𝑚𝑒𝑎𝑛 (8)

ROUGE-L gives an evaluation by automatically comparing generated summaries with human

reference summaries based on longest common subsequences (LCS). The LCS is between the generated
summaries and the human result summaries. If high similarity is shown then the summary system quality is
considered good [45]. Considering 𝑙(𝐶𝑖 , 𝑆𝑖𝑗 ) is the length of the LCS between two sentences, 𝑅𝑂𝑈𝐺𝐸𝐿 is
obtained by calculating the F-measure [40]. In (9)-(11) are used to calculate the metrics. 𝑅𝑙 is recall and 𝑃𝑙 is
precision of LCS. While 𝛽 is typically set to favor recall (𝛽 = 1.2).

𝑙(𝑐𝑖 ,𝑠𝑖𝑗 )
𝑅𝑙 = 𝑚𝑎𝑥𝑗 (9)
|𝑆𝑖𝑗 |

𝑙(𝑐𝑖 ,𝑠𝑖𝑗 )
𝑃𝑙 = 𝑚𝑎𝑥𝑗 (10)
|𝑐𝑖 |

(1+𝛽 2 )𝑅𝑙 ,𝑃𝑙

𝑅𝑂𝑈𝐺𝐸𝐿 (𝑐𝑖 , 𝑆𝑖 ) = (11)
𝑅𝑙 ⁡+⁡𝛽 2 𝑃𝑙

Consensus-based image description evaluation (CIDEr) gives measurements to the consensus

between candidate and reference texts using n-gram matching. Term frequency inverse document frequency
(TF-IDF) weighting is calculated for n-grams that are common in all texts [38]. The frequency of occurrence
of n-grams 𝜔𝑘 in the reference sentence for the candidate sentence 𝑐𝑖 is denoted by ℎ𝑘 (𝑠𝑖𝑗 ) or ℎ𝑘 (𝑐𝑖 ). 𝐶𝐼𝐷𝐸𝑟
calculates the TF-IDF weighting 𝑔𝑘 (𝑠𝑖𝑗 ) for each n-gram 𝜔𝑘 by using [40] (12). 𝐼 represents the set of all
images in the dataset and 𝛺 represents the vocabulary of all n-grams. The first term calculates the TF of each
n-gram 𝜔𝑘 , while the second term calculates the rarity of 𝜔𝑘 by using its IDF. To calculate the 𝐶𝐼𝐷𝐸𝑟𝑛 score
for n-grams of length 𝑛, we use the mean cosine similarity between candidate sentences and reference
sentences. considering precision and recall the calculation is as in (13). The vector 𝑔𝑛 (𝑐𝑖 ) represents all n-
grams of length n and is formed by 𝑔𝑘 (𝑐𝑖 ), while ‖𝑔𝑛 (𝑐𝑖 )‖ represents their magnitude. Likewise, for⁡𝑔𝑛 (𝑠𝑖𝑗 ).
Grammatical properties and richer semantics were captured by longer n-grams. The scores from n-grams of
various lengths were combined as (14). 𝑤𝑛 = 1/𝑁 is used for the uniform weights, with 4 as the value of 𝑁.

ℎ𝑘 (𝑠𝑖𝑗) |𝐼|
𝑔𝑘 (𝑠𝑖𝑗 ) = ∑ log (∑ ) (12)
𝜔𝑙 ∈Ω ℎ𝑙 (𝑠𝑖𝑗 ) 𝐼𝑝 𝜖⁡𝐼 min⁡(1,∑𝑞 ℎ𝑘 (𝑠𝑝𝑞 ))

1 𝑔𝑛 (𝑐𝑖 )⁡⋅⁡𝑔𝑛 (𝑆𝑖𝑗 )

𝐶𝐼𝐷𝐸𝑟𝑛 (𝑐𝑖 , 𝑆𝑖 ⁡) = ∑𝑗 (13)
𝑚 ‖𝑔𝑛 (𝑐𝑖 )‖‖𝑔𝑛 (𝑠𝑖𝑗 )‖

𝐶𝐼𝐷𝐸𝑟(𝑐𝑖 , 𝑆𝑖 ⁡) = ⁡ ∑𝑁
𝑛=1 𝑤𝑛 𝐶𝐼𝐷𝐸𝑟𝑛 (𝑐𝑖 , 𝑆𝑖 ) (14)

Semantic propositional image caption evaluation (SPICE) estimates text quality by converting
candidate and reference texts into semantic representations called “scene graphs” that encode objects,
attributes, and relationships found in the text [44]. In this evaluation, we first define the parsing captions’
subtask to scene graphs. We parse a caption 𝑐 into a scene graph, given a set of attribute types A, a set of object
classes C, and a set of relation types R, as (15) [46]. 𝑂(𝑐) ⁡ ⊆ 𝐶 is the set of objects named in 𝑐, 𝐸(𝑐) ⁡ ⊆
𝑂(𝑐) × 𝑅 × 𝑂(𝑐) is a hyper-edge set that represents the relationship between objects, and 𝐾(𝑐) ⁡ ⊆ 𝑂(𝑐) ⁡ × 𝐴
is the set of attributes related to the object. For the second step, we calculate the F-score. We view the scene
graph semantic relationships as a conjunction of logical propositions or tuples, to compare how closely two
scene graphs, resemble one another. So, we have a function T that reads the scene graph and returns a logical

Int J Artif Intell, Vol. 13, No. 1, March 2024: 23-34

Int J Artif Intell ISSN: 2252-8938  29

tuple as (16). Each tuple consists of one, two, or three components that, accordingly, represent the objects,
attributes, and relations. We define the binary matching operator ⊗ as the function that returns matching tuples
in two scene graphs by looking at the semantic propositions in the scene graph as a set of tuples. Next, we
define 𝑆𝑃𝐼𝐶𝐸, recall 𝑅, and precision 𝑃 as (17)-(19).

𝐺(𝑐) = 〈𝑂(𝑐), 𝐸(𝑐), 𝐾(𝑐)〉 (15)

𝑇(𝐺(𝑐) ≜ 𝑂(𝑐) ⁡ ∪ 𝐸(𝑐) ⁡ ∪ 𝐾(𝑐) (16)

|𝑇(𝐺(𝑐))⁡⊗⁡𝑇(𝐺(𝑆))|
𝑃(𝑐, 𝑆) = ⁡ |𝑇(𝐺(𝑐))|
(17)

|𝑇(𝐺(𝑐))⁡⊗⁡𝑇(𝐺(𝑆))|
𝑅(𝑐, 𝑆) = ⁡ |𝑇(𝐺(𝑆))|
(18)

2⁡⋅⁡𝑃(𝑐,𝑆)⁡⋅⁡𝑅(𝑐,𝑆)
𝑆𝑃𝐼𝐶𝐸(𝑐, 𝑆) = 𝐹1 (𝑐, 𝑆) = ⁡ (19)
𝑃(𝑐,𝑆)+𝑅(𝑐,𝑆)

Of all the metrics used, namely BLEU-n, METEOR, ROGUE-L, CIDEr and SPICE, we selected four
evaluation metrics that are used in almost all literature: BLEU-4, ROUGE-L, CIDEr, and METEOR. Table 4
shows the average value obtained in the evaluation metric of the transformer and attention architecture used in
the literature discussed. However, calculations in Table 4 only include literature that contains c5 and c40 scores.
It can be seen from the table that the transformer model gets a higher score for each evaluation metric than the
vanilla attention mechanism model.

Table 4. Average score for each evaluation metric

BLEU-4 ROUGE-L CIDEr METEOR
Evaluation metric/Architecture
C5 C40 C5 C40 C5 C40 C5 C40
Transformer 39.68 72.00 59.27 74.58 129.49 131.59 29.32 38.76
Attention 36.41 66.95 56.94 71.98 116.06 118.14 27.48 36.53

3.3. General concept

At this stage of analysis, the steps for modeling the image captioning model that are commonly found
are formulated. The general model formulation factors are taken from the dataset, preprocessing, architecture,
and methods side, as well as the evaluation used. Figure 1 shows a generic flow chart based on the general
modeling of the literature review.
Many of the datasets used are open-source. The dataset is available along with the distribution of the
dataset for image captioning, such as Flickr30k with data totaling 30 thousand images, Flickr8k with data
totaling 8 thousand images, MS COCO 180 thousand images, and Conceptual 3.3 million images. To see the
distribution of the dataset, the number of images and the literature can be seen in Table 3.
Preprocessing is mostly done by following the steps of Karpathy [39]. To perform the preprocessing
stage of the MS COCO dataset, you can follow the available source code [47]. Karpathy's preprocessing stage
includes tokenizing, lowercase the text, then changing the word to "UNK" (unknown) or deleting words whose
frequency is less than 5. In addition, some literatures also set varying caption lengths for MS COCO and
Flickr30k [12], [20]. These words are then represented using GloVe or one-hot-encoding [2], [25].
The architecture and methods studied show various transformer and attention models. Many models
rely on encoder and decoder frameworks because they are considered flexible [48]. The flexibility is not only
in the designing the model architecture, but also the flexibility in implementing such architecture to different
domain, for instance, molecular image captioning [49]. The role of the encoder is to extract features from the
input image. While the decoder is useful for generating grammatically appropriate words. In this study, the
most widely used encoder is the CNN variation model, while the most widely used decoders are LSTM, CNN,
and RNN [2], [5], [7], [8], [42]. The method used is a modification of transformers and attention. The discussion
of the method can be seen in section 3.1. The methods studied were obtained from 36 literatures. These methods
are evaluated using the original dataset or from Karpathy [8].
The evaluations that are widely used are the BLEU, METEOR, ROUGE-L, CIDEr, and SPICE evaluation
metrics. The evaluation score is widely used to evaluate the results of image captioning. To evaluate much of the
literature we reviewed, we used the source code available from MS COCO [50] to calculate the evaluation metric.
From all the literature that we reviewed, the evaluation metrics that are often used are BLEU-4, Rouge-L, CIDEr,
and METEOR which are popular and known to have a strong correlation with human judgment [51], [52].

A study on attention-based deep learning architecture model for image captioning (Dhomas Hatta Fudholi)
30  ISSN: 2252-8938

Figure 1. General flow diagram of image captioning model and concept

4. CONCLUSION
From this study on attention-based deep learning architecture model for image captioning, 36
literatures were found. With all models are attention-based architecture, half of the works listed in this study
uses transformer. Five types of evaluation metrics were used across the works: BLEU, CIDEr, METEOR,
ROUGE-L, and SPICE. BLEU still become the most used evaluation metrics for image captioning. From the
analysis, it is known that on average, the transformer model obtains higher evaluation metric score at
BLEU-4, CIDEr, ROUGE-L, and METEOR than the works with vanilla attention mechanism model. In the
Indonesian language domain, as one example of a low resource language, only few works are found and most
of them still rely on the common MSCOCO dataset as the base. However, the is an effort to create a novel
dataset which is great to capture local culture in the caption. Finally, this study provides a foundation for the
future development of the image captioning model and presents a general understanding of the process of
developing image captions, including a representation of the process in low resource languages.

ACKNOWLEDGEMENT
The study is sponsored by SAME (Scheme for Academic Mobility and Exchange) 2022 program
(Decree No. 3253/E4/DT.04.03/2022) from Directorate of Resources, Directorate General of Higher Education,
Research and Technology, Ministry of Education, Culture, Research and Technology, Republic of Indonesia.

APPENDIX

Table 1. Literature review result

Literature Architecture Method Result on MSCOCO online testing server
[1] Transformer Image transformer BLEU-1,4(c5, c40): (81.2, 95.4), (39.6, 71.5); METEOR(c5, c40): (29.1,
38.4); ROUGE-L(c5, c40): (59.2, 74.5); CIDEr(c5, c40): (127.4, 129.6);
SPICE(c5, c40): -
[4] Transformer Meshed-memory BLEU-1,2,3,4(c5, c40): (81.6, 96.0), (66.4, 90.8), (51.8, 82.7), (39.7, 72.8);
transformer METEOR(c5, c40): (29.4, 39.0); ROUGE-L(c5, c40): (59.2, 74.8);
CIDEr(c5, c40): (129.3, 132.1); SPICE(c5, c40): -
[5] Attention Long short-term with BLEU-1,2,3,4: - ; METEOR: 23.4; ROUGE-L: - ; CIDEr: 88.3; SPICE:
pointing 16.6

Int J Artif Intell, Vol. 13, No. 1, March 2024: 23-34

Int J Artif Intell ISSN: 2252-8938  31

Table 1. Literature review result (continue)

Literature Architecture Method Result on MSCOCO online testing server
[6] Attention Hierarchy parsing BLEU-1,2,3,4(c5, c40): (81.6, 95.9), (66.2, 90.4), (51.5, 81.6), (39.3, 71.0);
METEOR(c5, c40): (28.8, 38.1); ROUGE-L(c5, c40): (59.0, 74.1);
CIDEr(c5, c40): (127.9, 130.2); SPICE(c5, c40): -
[2] Transformer Multimodal Transformer BLEU-1,2,3,4(c5, c40): (81.7, 95.6), (66.8, 90.5), (52.4, 82.4), (40.4, 72.2);
with multi-view Visual METEOR(c5, c40): (29.4, 38.9); ROUGE-L(c5, c40): (59.6, 75.0);
representation CIDEr(c5, c40): (130.0, 130.9); SPICE(c5, c40): -
[7] Attention Hierarchical attention BLEU-1,2,4(c5, c40): (80.4, 94.5), (63.8, 87.7), (36.5, 66.8); METEOR(c5,
network c40): (27.4, 36.1); ROUGE-L(c5, c40): (57.3, 71.9); CIDEr(c5, c40):
(115.2, 118.2); SPICE(c5, c40): -
[8] Attention Attention on attention BLEU-1,2,3,4(c5, c40): (81.0, 95.0), (65.8, 89.6), (51.4, 81.3), (39.4, 71.2);
METEOR(c5, c40): (29.1, 38.5); ROUGE-L(c5, c40): (58.9, 74.5);
CIDEr(c5, c40): (126.9, 129.6); SPICE(c5, c40): -
[9] Transformer Boosted transformer BLEU-1,2,3,4(c5, c40): (80.5, 94.7), (65.2, 88.8), (50.6, 80.1), (38.6, 69.6);
METEOR(c5, c40): (28.8, 37.9); ROUGE-L(c5, c40): (58.5, 73.5);
CIDEr(c5, c40): (125.0, 126.8); SPICE(c5, c40): -
[10] Transformer Object relation BLEU-1,4: 80.5, 38.6; METEOR: 28.7; ROUGE-L: 58.4; CIDEr: 128.3;
transformer SPICE: 22.6
[11] Transformer EnTangled transformer BLEU-1,2,3,4(c5, c40): (81.2, 95.0), (65.5, 89.0), (50.9, 80.4), (38.9, 70.2);
METEOR(c5, c40): (28.6, 38.0); ROUGE-L(c5, c40): (58.6, 73.9);
CIDEr(c5, c40): (122.1, 124.4); SPICE(c5, c40): -
[12] Attention Scene graph auto- BLEU-4(c5, c40): (38.5, 69.7); METEOR(c5, c40): (28.2, 37.2); ROUGE-
encoder L(c5, c40): (58.6, 73.6); CIDEr(c5, c40): (123.8, 126.5); SPICE(c5, c40): -
[13] Transformer TransResNet BLEU-1,4: 79.3, 36.4; METEOR: -; ROUGE-L: 57.5; CIDEr: 124.0;
SPICE: 21.2
[14] Transformer RNN and transformer BLEU-1,2,3,4: - ; METEOR: - ; ROUGE-L: 0.336; CIDEr: 1.676; SPICE:
0.257
[15] Attention Recurrent fusion BLEU-1,2,3,4(c5, c40): (80.4, 95.0), (64.9, 89.3), (50.1, 90.1), (38.0, 69.2);
network METEOR(c5, c40): (28.2, 37.2); ROUGE-L(c5, c40): (58.2, 73.1);
CIDEr(c5, c40): (122.9, 125.1); SPICE(c5, c40): -
[16] Attention unsupervised learning BLEU-1,2,3,4: 58.9, 40.3, 27.0, 18.6; METEOR: 18.6; ROUGE-L: 43.1;
CIDEr: 54.9; SPICE: 11.1
[17] Attention Graph convolutional BLEU-2,3,4(c5, c40): (65.5, 89.3), (50.8, 80.3), (38.7, 68.7); METEOR(c5,
networks-LSTM c40): (28.5, 37.6); ROUGE-L(c5, c40): (58.5, 73.4); CIDEr(c5, c40):
(125.3, 126.5); SPICE(c5, c40): -
[18] Transformer Stacked attention BLEU-1,2,3,4: 73.0, 56.9, 43.6, 33.3; METEOR: -; ROUGE-L: 54.8;
modules CIDEr: 108.1; SPICE: -
[19] Attention Bottom-up and top- BLEU-1,2,3,4(c5, c40): (80.2, 95.2), (64.1, 88.8), (49.1, 79.4), (36.9, 68.5);
down attention METEOR(c5, c40): (27.6, 36.7); ROUGE-L(c5, c40): (57.1, 72.4);
CIDEr(c5, c40): (117.9, 120.5); SPICE(c5, c40): (21.5, 71.5)
[20] Attention Adaptive attention BLEU-1,2,3,4(c5, c40): (0.748, 0.920), (0.584, 0.845), (0.444, 0.744),
model with visual (0.336, 0.637); METEOR(c5, c40): (0.264, 0.359); ROUGE-L(c5, c40):
sentinel (0.550, 0.705); CIDEr(c5, c40): (1.042, 1.059); SPICE(c5, c40): -
[21] Attention Spatial and channel-wise BLEU-1,2,3,4(c5, c40): (71.2, 89.4), (54.2, 80.2), (40.4, 69.1), (30.2, 57.9);
attentions in a CNN METEOR(c5, c40): (24.4, 33.1); ROUGE-L(c5, c40): (52.4, 67.4);
CIDEr(c5, c40): (91.2, 92.1); SPICE(c5, c40): -
[22] Attention Policy gradient BLEU-1,2,3,4: 0.743, 0.578, 0.433, 0.322; METEOR: 0.251; ROUGE-L:
optimization of SPIDEr 0.544; CIDEr: 1.000; SPICE: -
[23] Attention Convolutional image BLEU-1,2,3,4(c5, c40): (0.715, 0.896), (0.545, 0.805), (0.408, 0.693),
captioning (0.304, 0.582); METEOR(c5, c40): (0.246, 0.333); ROUGE-L(c5, c40):
(0.525, 0.673); CIDEr(c5, c40): (0.910, 0.914); SPICE(c5, c40): -
[24] Attention Attention correctness BLEU-3,4: 38.0, 28.1; METEOR: 23.01; ROUGE-L: -; CIDEr: -; SPICE: -
[25] Attention Long short-term BLEU-1,2,3,4(c5, c40): (78.7, 93.7), (62.7, 86.7), (47.6, 76.5), (35.6, 65.2);
memory with attributes METEOR(c5, c40): (27.0, 35.4); ROUGE-L(c5, c40): (56.4, 70.5);
CIDEr(c5, c40): (116.0, 118.0); SPICE(c5, c40): -
[26] Transformer Dual-level collaborative BLEU-1,2,3,4(c5, c40): (82.4, 96.6), (67.4, 91.7), (52.8, 83.8), (40.6, 74.0);
transformer METEOR(c5, c40): (29.8, 39.6); ROUGE-L(c5, c40): (59.8, 75.3);
CIDEr(c5, c40): (133.3, 135.4); SPICE: -
[27] Transformer Global enhanced BLEU-1,2,3,4(c5, c40): (81.6, 96.1), (66.5, 90.9), (51.9, 82.8), (39.7, 72.9);
transformer METEOR(c5, c40): (29.4, 38.8); ROUGE-L(c5, c40): (59.1, 74.4);
CIDEr(c5, c40): (130.3, 132.5); SPICE: -
[28] Transformer Caption transformer BLEU-1,2,3,4(c5, c40): (81.8, 95.0), (66.5, 89.4), (51.8, 80.9), (39.5, 70.8);
METEOR(c5, c40): (29.1, 38.3); ROUGE-L(c5, c40): (59.2, 74.4);
CIDEr(c5, c40): (125.4, 127.3); SPICE: -
[29] Transformer Semi-autoregressive BLEU-1,2,3,4(c5, c40): (80.3, 94.5), (64.4, 87.9), (49.2, 78.2), (37.0, 67.2);
transformer METEOR(c5, c40): (28.2, 37.0); ROUGE-L(c5, c40): (57.8, 72.6);
CIDEr(c5, c40): (121.5, 124.1); SPICE: -
[30] Transformer Spatial and scale-aware BLEU-1,2,3,4(c5, c40): (82.2, 96.5), (67.0, 91.4), (52.4, 83.3), (40.1, 73.5);
transformer METEOR(c5, c40): (29.6, 39.3); ROUGE-L(c5, c40): (59.5, 75.0);
CIDEr(c5, c40): (132.6, 135.0); SPICE: -
[31] Transformer Faster regions with BLEU-1,4(c5, c40): (82.0, 96.7), (40.1, 73.2); METEOR(c5, c40): (29.8,
convolutional neural 39.5); ROUGE-L(c5, c40): (59.9, 75.2); CIDEr(c5, c40): (129.9, 132.8);
networks SPICE: -

A study on attention-based deep learning architecture model for image captioning (Dhomas Hatta Fudholi)
32  ISSN: 2252-8938

Table 1. Literature review result (continue)

Literature Architecture Method Result on MSCOCO online testing server
[32] Transformer Attention-reinforced BLEU-1,2,3,4(c5, c40): (83.4, 97.4), (68.8, 93.0), (54.3, 85.6), (42.0, 76.1);
transformer METEOR(c5, c40): (30.6, 40.4); ROUGE-L(c5, c40): (60.8, 76.4);
CIDEr(c5, c40): (138.5, 140.5); SPICE: -
[33] Attention Prophet attention BLEU-1,2,3,4(c5, c40): (81.8, 96.3), (66.5, 91.2), (51.9, 83.2), (39.8, 73.3);
METEOR(c5, c40): (29.6, 39.3); ROUGE-L(c5, c40): (59.4, 75.1);
CIDEr(c5, c40): (130.4, 133.7); SPICE: -

REFERENCES
[1] S. He, W. Liao, H. R. Tavakoli, M. Yang, B. Rosenhahn, and N. Pugeault, “Image captioning through image transformer,” Lecture
Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 12625
LNCS, pp. 153–169, 2021, doi: 10.1007/978-3-030-69538-5_10.
[2] J. Yu, J. Li, Z. Yu, and Q. Huang, “Multimodal transformer with multi-view visual representation for image captioning,” IEEE
Transactions on Circuits and Systems for Video Technology, vol. 30, no. 12, pp. 4467–4480, 2020, doi: 10.1109/TCSVT.2019.2947482.
[3] A. Vaswani et al., “Attention is all you need,” in Proceedings of the 31st Conference on Neural Information Processing Systems, Dec.
2017, pp. 5998–6008, doi: 10.48550/arXiv.1706.03762.
[4] M. Cornia, M. Stefanini, L. Baraldi, and R. Cucchiara, “Meshed-memory transformer for image captioning,” Proceedings of the IEEE
Computer Society Conference on Computer Vision and Pattern Recognition, pp. 10575–10584, 2020, doi: 10.1109/CVPR42600.
2020.01059.
[5] Y. Li, T. Yao, Y. Pan, H. Chao, and T. Mei, “Pointing novel objects in image captioning,” Proceedings of the IEEE Computer Society
Conference on Computer Vision and Pattern Recognition, vol. 2019-June, pp. 12489–12498, 2019, doi: 10.1109/CVPR.2019.01278.
[6] T. Yao, Y. Pan, Y. Li, and T. Mei, “Hierarchy parsing for image captioning,” Proceedings of the IEEE International Conference on
Computer Vision, vol. 2019-Oct., pp. 2621–2629, 2019, doi: 10.1109/ICCV.2019.00271.
[7] W. Wang, Z. Chen, and H. Hu, “Hierarchical attention network for image captioning,” 33rd AAAI Conference on Artificial Intelligence,
AAAI 2019, 31st Innovative Applications of Artificial Intelligence Conference, IAAI 2019 and the 9th AAAI Symposium on Educational
Advances in Artificial Intelligence, EAAI 2019, pp. 8957–8964, 2019, doi: 10.1609/aaai.v33i01.33018957.
[8] L. Huang, W. Wang, J. Chen, and X. Y. Wei, “Attention on attention for image captioning,” Proceedings of the IEEE International
Conference on Computer Vision, vol. 2019-Oct., pp. 4633–4642, 2019, doi: 10.1109/ICCV.2019.00473.
[9] J. Li, P. Yao, L. Guo, and W. Zhang, “Boosted transformer for image captioning,” Applied Sciences (Switzerland), vol. 9, no. 16,
pp. 1–15, 2019, doi: 10.3390/app9163260.
[10] S. Herdade, A. Kappeler, K. Boakye, and J. Soares, “Image captioning: Transforming objects into words,” Advances in Neural
Information Processing Systems, vol. 32, no. NeurIPS, pp. 1–11, 2019.
[11] G. Li, L. Zhu, P. Liu, and Y. Yang, “Entangled transformer for image captioning,” Proceedings of the IEEE International Conference
on Computer Vision, vol. 2019-Oct., no. C, pp. 8927–8936, 2019, doi: 10.1109/ICCV.2019.00902.
[12] X. Yang, K. Tang, H. Zhang, and J. Cai, “Auto-encoding scene graphs for image captioning,” Proceedings of the IEEE Computer
Society Conference on Computer Vision and Pattern Recognition, vol. 2019-June, pp. 10677–10686, 2019, doi: 10.1109/CVPR.
2019.01094.
[13] K. Shuster, S. Humeau, H. Hu, A. Bordes, and J. Weston, “Engaging image captioning via personality,” Proceedings of the IEEE
Computer Society Conference on Computer Vision and Pattern Recognition, vol. 2019-June, pp. 12508–12518, 2019, doi: 10.1109/
CVPR.2019.01280.
[14] P. Sharma, N. Ding, S. Goodman, and R. Soricut, “Conceptual captions: A cleaned, hypernymed, image alt-text dataset for automatic
image captioning,” ACL 2018-56th Annual Meeting of the Association for Computational Linguistics, Proceedings of the Conference
(Long Papers), vol. 1, pp. 2556–2565, 2018, doi: 10.18653/v1/p18-1238.
[15] W. Jiang, L. Ma, Y. G. Jiang, W. Liu, and T. Zhang, “Recurrent fusion network for image captioning,” Lecture Notes in Computer
Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11206 LNCS,
pp. 510–526, 2018, doi: 10.1007/978-3-030-01216-8_31.
[16] Y. Feng, L. Ma, W. Liu, and J. Luo, “Unsupervised image captioning,” Proceedings of the IEEE Computer Society Conference on
Computer Vision and Pattern Recognition, vol. 2019-June, pp. 4120–4129, 2019, doi: 10.1109/CVPR.2019.00425.
[17] T. Yao, Y. Pan, Y. Li, and T. Mei, “Exploring visual relationship for image captioning,” Lecture Notes in Computer Science (including
subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11218 LNCS, pp. 711–727, 2018,
doi: 10.1007/978-3-030-01264-9_42.
[18] X. Zhu, L. Li, J. Liu, H. Peng, and X. Niu, “Captioning transformer with stacked attention modules,” Applied Sciences (Switzerland),
vol. 8, no. 5, 2018, doi: 10.3390/app8050739.
[19] P. Anderson et al., “Bottom-up and top-down attention for image captioning and visual question answering,” Proceedings of the IEEE
Computer Society Conference on Computer Vision and Pattern Recognition, pp. 6077–6086, 2018, doi: 10.1109/CVPR.2018.00636.
[20] J. Lu, C. Xiong, D. Parikh, and R. Socher, “Knowing when to look: Adaptive attention via a visual sentinel for image captioning,”
Proceedings-30th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2017, vol. 2017-Jan., pp. 3242–3250, 2017,
doi: 10.1109/CVPR.2017.345.
[21] L. Chen et al., “SCA-CNN: Spatial and channel-wise attention in convolutional networks for image captioning,” Proceedings-30th
IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2017, vol. 2017-Jan., pp. 6298–6306, 2017, doi: 10.1109/CVPR.
2017.667.
[22] S. Liu, Z. Zhu, N. Ye, S. Guadarrama, and K. Murphy, “Improved image captioning via policy gradient optimization of SPIDEr,”
Proceedings of the IEEE International Conference on Computer Vision, vol. 2017-Oct., pp. 873–881, 2017, doi: 10.1109/ICCV.
2017.100.
[23] J. Aneja, A. Deshpande, and A. G. Schwing, “Convolutional image captioning,” Proceedings of the IEEE Computer Society Conference
on Computer Vision and Pattern Recognition, pp. 5561–5570, 2018, doi: 10.1109/CVPR.2018.00583.
[24] C. Liu, J. Mao, F. Sha, and A. Yuille, “Attention correctness in neural image captioning,” 31st AAAI Conference on Artificial
Intelligence, AAAI 2017, pp. 4176–4182, 2017, doi: 10.1609/aaai.v31i1.11197.
[25] T. Yao, Y. Pan, Y. Li, Z. Qiu, and T. Mei, “Boosting image captioning with attributes,” Proceedings of the IEEE International
Conference on Computer Vision, vol. 2017-Octob, pp. 4904–4912, 2017, doi: 10.1109/ICCV.2017.524.

Int J Artif Intell, Vol. 13, No. 1, March 2024: 23-34

Int J Artif Intell ISSN: 2252-8938  33

[26] Y. Luo et al., “Dual-level collaborative transformer for image captioning,” 35th AAAI Conference on Artificial Intelligence, AAAI 2021,
vol. 3B, no. 3, pp. 2286–2293, doi: 10.1609/aaai.v35i3.16328.
[27] J. Ji et al., “Improving image captioning by leveraging intra- and inter-layer global representation in transformer network,” 35th AAAI
Conference on Artificial Intelligence, AAAI 2021, vol. 2B, pp. 1655–1663, 2021, doi: 10.1609/aaai.v35i2.16258.
[28] W. Liu, S. Chen, L. Guo, X. Zhu, and J. Liu, “CPTR: Full transformer network for image captioning,” Computer Science > Computer
Vision and Pattern Recognition, pp. 1–5, 2021, doi: 10.48550/arXiv.2101.10804.
[29] Y. Zhou, Y. Zhang, Z. Hu, and M. Wang, “Semi-autoregressive transformer for image captioning,” Proceedings of the IEEE
International Conference on Computer Vision, vol. 2021-Oct., pp. 3132–3136, 2021, doi: 10.1109/ICCVW54120.2021.00350.
[30] P. Zeng, H. Zhang, J. Song, and L. Gao, “S2 transformer for image captioning,” IJCAI International Joint Conference on Artificial
Intelligence, pp. 1608–1614, 2022, doi: 10.24963/ijcai.2022/224.
[31] X. Yang, Y. Liu, and X. Wang, “ReFormer: the relational transformer for image captioning,” in Proceedings of the 30th ACM
International Conference on Multimedia, Oct. 2022, pp. 5398–5406, doi: 10.1145/3503161.3548409.
[32] Z. Wang, S. Shi, Z. Zhai, Y. Wu, and R. Yang, “ArCo: Attention-reinforced transformer with contrastive learning for image captioning,”
Image and Vision Computing, vol. 128, p. 104570, Dec. 2022, doi: 10.1016/j.imavis.2022.104570.
[33] F. Liu, X. Ren, X. Wu, W. Fan, Y. Zou, and X. Sun, “Prophet attention: Predicting attention with future attention for image captioning,”
Computer Science > Computer Vision and Pattern Recognition, 2022, doi: 10.48550/arxiv.2210.10914.
[34] M. R. S. Mahadi, A. Arifianto, and K. N. Ramadhani, “Adaptive attention generation for Indonesian image captioning,” in 2020 8th
International Conference on Information and Communication Technology (ICoICT), Jun. 2020, pp. 1–6, doi: 10.1109/ICoICT49345.
2020.9166244.
[35] D. H. Fudholi et al., “Image captioning with attention for smart local tourism using efficientNet,” IOP Conference Series: Materials
Science and Engineering, vol. 1077, no. 1, p. 012038, 2021, doi: 10.1088/1757-899x/1077/1/012038.
[36] R. Mulyawan, A. Sunyoto, and A. H. Muhammad, “Automatic Indonesian image captioning using CNN and transformer-based model
approach,” ICOIACT 2022-5th International Conference on Information and Communications Technology: A New Way to Make AI
Useful for Everyone in the New Normal Era, Proceeding, pp. 355–360, 2022, doi: 10.1109/ICOIACT55506.2022.9971855.
[37] U. A. A. Al-Faruq and D. H. Fudholi, “EfficientNet-Transformer for image captioning in Bahasa,” Vii International Conference “Safety
Problems of Civil Engineering Critical Infrastructures” (Spceci2021), vol. 2701, p. 020037, 2023, doi: 10.1063/5.0118155.
[38] T.-Y. Lin et al., “Microsoft COCO: common objects in context,” in In: European conference on computer vision, May 2014,
pp. 740–755, doi: 10.1007/978-3-319-10602-1_48.
[39] A. Karpathy and L. Fei-Fei, “Deep visual-semantic alignments for generating image descriptions,” IEEE Transactions on Pattern
Analysis and Machine Intelligence, vol. 39, no. 4, pp. 664–676, 2017, doi: 10.1109/TPAMI.2016.2598339.
[40] X. Chen et al., “Microsoft COCO captions: Data collection and evaluation server,” Computer Science > Computer Vision and Pattern
Recognition, 2015, doi: 10.48550/arXiv.1504.00325.
[41] P. Young, A. Lai, M. Hodosh, and J. Hockenmaier, “From image descriptions to visual denotations: New similarity metrics for semantic
inference over event descriptions,” Transactions of the Association for Computational Linguistics, vol. 2, pp. 67–78, Dec. 2014,
doi: 10.1162/tacl_a_00166.
[42] S. He and Y. Lu, “A modularized architecture of multi-branch convolutional neural network for image captioning,” Electronics, vol. 8,
no. 12, p. 1417, Nov. 2019, doi: 10.3390/electronics8121417.
[43] Y. Cui, G. Yang, A. Veit, X. Huang, and S. Belongie, “Learning to evaluate image captioning,” in 2018 IEEE/CVF Conference on
Computer Vision and Pattern Recognition, Jun. 2018, pp. 5804–5812, doi: 10.1109/CVPR.2018.00608.
[44] N. Sharif, L. White, M. Bennamoun, and S. A. A. Shah, “NNEval: Neural network based evaluation metric for image captioning,” in
ECCV 2018: Computer Vision – ECCV 2018, 2018, pp. 39–55, doi: 10.1007/978-3-030-01237-3_3.
[45] A. Cohan and N. Goharian, “Revisiting summarization evaluation for scientific articles,” Proceedings of the 10th International
Conference on Language Resources and Evaluation, LREC 2016, pp. 806–813, 2016, doi: 10.48550/arXiv.1604.00400.
[46] P. Anderson, B. Fernando, M. Johnson, and S. Gould, “SPICE: Semantic propositional image caption evaluation,” Lecture Notes in
Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 9909 LNCS,
pp. 382–398, 2016, doi: 10.1007/978-3-319-46454-1_24.
[47] Karpathy, “GitHub-karpathy/neuraltalk: NeuralTalk is a Python+numpy project for learning multimodal recurrent neural networks that
describe images with sentences,” GitHub, 2015. https://github.com/karpathy/neuraltalk (accessed May 27, 2023).
[48] R. Staniute and D. Šešok, “A systematic literature review on image captioning,” Applied Sciences (Switzerland), vol. 9, no. 10,
pp. 1–20, 2019, doi: 10.3390/app9102024.
[49] J. Yi et al., “MICER: a pre-trained encoder-decoder architecture for molecular image captioning,” Bioinformatics (Oxford, England),
vol. 38, no. 19, pp. 4562–4572, 2022, doi: 10.1093/bioinformatics/btac545.
[50] Tylin, “GitHub-tylin/coco-caption,” GitHub, 2018. https://github.com/tylin/coco-caption (accessed May 27, 2023).
[51] N. Aafaq, A. Mian, W. Liu, S. Z. Gilani, and M. Shah, “Video description: A survey of methods, datasets, and evaluation metrics,”
ACM Computing Surveys, vol. 52, no. 6, pp. 1–37, 2019, doi: 10.1145/3355390.
[52] Y. Graham, “Re-evaluating automatic summarization with BLEU and 192 shades of ROUGE,” Conference Proceedings-EMNLP 2015:
Conference on Empirical Methods in Natural Language Processing, pp. 128–137, 2015, doi: 10.18653/v1/d15-1013.

BIOGRAPHIES OF AUTHORS

Dhomas Hatta Fudholi is an Associate Professor at the Department of

Informatics, Universitas Islam Indonesia, Yogyakarta, Indonesia. He earned his Ph.D. in
Computer Science and IT in 2016 from La Trobe University, Melbourne, Australia. He
explores data science and deep learning methods to support knowledge base development
and various useful applications. He can be contacted at email: [email protected].

A study on attention-based deep learning architecture model for image captioning (Dhomas Hatta Fudholi)
34  ISSN: 2252-8938

Umar Abdul Aziz Al-Faruq is a freshgraduate from Islamic University of

Indonesia. He earned his bachelor’s degree in informatics. He gained a strong foundation in
computer science and programming. Some of his projects are related to Artificial Intelligent
and web development. He can be contacted at email: [email protected].

Royan Abida Nur Nayoan is a Research and Development Engineer in the field
of Artificial Intelligence, specializing in computer vision and natural language processing.
With a Master's degree in Informatics from Universitas Islam Indonesia, she is currently
focusing on developing new AI applications to enhance company products. She has
contributed to the field of AI through research publications and presentations. With a passion
for AI, she is dedicated to staying up-to-date with the latest advancements in the field. She
can be contacted at email: [email protected].

Annisa Zahra is currently a master’s student studying Artificial Intelligence at

Universitas Gadjah Mada, Indonesia. She obtained her bachelor’s degree in informatics in
2021 from Universitas Islam Indonesia, Indonesia. She has done several machine learning
projects in the domain of natural language processing. She can be contacted at email:
[email protected].

Int J Artif Intell, Vol. 13, No. 1, March 2024: 23-34