Post 1:
Veto Bishop
Research Design
A very fundamental tool in scientific research methodology (Apuke, 2017). Researchers have the follois
wing three choices when it comes to designing their methodology: quantitative, qualitative, hybrid. Butor
the choice that is made inextricably linked is to the research design. The research design is the blueprin
t from the entire study. The design the structure, strategy, and plan the research question will is on how
be answered (Akhtar, 2016). answer To my research question, I will employing the quantitative approbe
ach. This approach has been chosen because the model will be evaluated against mathematical metric
s. It is a possibility that the output could evaluated using humans, but time does not permit such be an a
pproach. So, I could have opted for a hybrid approach. However, for the sake the time allotted the of qu
antitative approach employed. The remaining section will explain each step research plan. Iwill be of my
t will start discussing the dataset that will used. by be
Dataset
To conduct the experiments picture caption creation, a variein
ty of datasets are available. I will use the Flickr30k Dataset for
English Caption Generation. For Spanish, MS-COCO-
ES will used which created from the original -be is MS
COCO dataset. This subs composed 20,000 captions et is of of
4,000 images captions per image) machine-(5 in
translated Spanish and original English that have been used al
ong with the main subset to carry out. The size the photos, tof
he number captions of per image, the structure the captionsof
, the number images all vary amongst the designed datand of
asets.
Methodology
The image captions (word sequences) are encoded using a recurrent neural network the proposed min
odel, while the image characteristics are encoded using a convolutional neural network. A text feature d
ecoder then receives the encoded image and guesses the caption word by word. The model used attenti
on to concentrate on the most crucial element the image of as it came with each phrase for the captioup
n. Figure 1 depicts the architecture the suggested model, which principally uses the encoder-of
decoder mechanism.
Figure 1
Model Overview
Encoder: Image Features
In the first stage, I pre-process the raw images using transfer learning and a -CNN
based network that has already been trained. This procedure takes input photos and generates encoded
image vectors that capture the key elements the input image. I utilized EfficientNetB3 (Tan & Le, of 201
9) and the pre-
