To efficiently train the large dataset using the 2D CNN model with good accuracy
To efficiently train the few shot images using the channel boost CNN model with good accuracy
To achieve the proper cleansing of the input image using the hybrid model which includes CB-CNN and 2D CNN
1. Few-Shot learning for enhancing dataset cleaning
1. Z. Guo et al., "Few-shot Fish Image Generation and Classification," Global Oceans 2020: Singapore – U.S. Gulf Coast, Biloxi, MS, USA, 2020, pp. 1-6, doi: 10.1109/IEEECONF38699.2020.9389005.
2. M. J. Lee and J. So, "Metric-Based Learning for Nearest-Neighbor Few-Shot Image Classification," 2021 International Conference on Information Networking (ICOIN), Jeju Island, Korea (South), 2021, pp. 460-464, doi:
10.1109/ICOIN50884.2021.9333850.
3. E. Patsiouras, A. Tefas and I. Pitas, "Few-shot Image Recognition for UAV Sports Cinematography," 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), Seattle, WA, USA,
2020, pp. 965-969, doi: 10.1109/CVPRW50498.2020.00127.
References
The process of data cleaning is crucial for enhancing the
quality and reliability of datasets used in machine learning models.
Traditional methods often struggle with the challenges posed by
noisy, incomplete, or erroneous data. This study introduces a novel
approach leveraging few-shot learning models to enhance the
dataset cleaning process. Few-shot learning is employed to handle
data with limited labeled examples, aiding in the identification and
correction of inconsistencies and outliers within datasets.
This research investigates the performance of a few-shot
learning model in comparison to established clustering methods
such as Support Vector Machines (SVM) and K-Nearest Neighbors
(KNN) for the task of dataset cleaning. The few-shot learning
model’s ability to generalize from limited labeled instances and
adapt to diverse and noisy data is a focus of this study.
Additionally, the study evaluates and compares the efficiency,
accuracy, and computational demands of the few-shot learning
approach against SVM and KNN algorithms.
Abstract
Introduction
Objective Block Diagram
The efficient identification of few-shot image enhancement
via channel gapping, coupled with proper image augmentation
using trained data features, marks a significant leap in image
processing. Addressing overfitting and underfitting challenges is
crucial for model reliability, ensuring a balance between
complexity and adaptability. Achieving a hybrid model for
enhancing few-shot images emerges as a promising strategy,
combining various techniques to optimize image enhancement with
limited training data. These integrated approaches pave the way for
robust and adaptable solutions in the realm of image enhancement
Conclusions
This project embarks on a comparative analysis between the
few-shot learning approach and determine the strengths,
weaknesses, and overall performance established clustering
models, specifically Support Vector Machines (SVM) and K-
Nearest Neighbors (KNN). The comparative assessment aims to
evaluate the efficacy and efficiency of the few-shot learning model
in the context of dataset cleaning. It seeks to f the few-shot
learning approach in contrast to traditional clustering methods in
handling and rectifying noisy or inconsistent data.
The research endeavors to provide a comprehensive analysis
of these methodologies by conducting experiments on diverse
datasets with varying complexities and types of noise. The
evaluation encompasses multiple metrics such as accuracy,
precision, recall, F1-score, and computational efficiency to gauge
the performance of each approach.
Through this study, a deep understanding of the potential of few-
shot learning in improving the dataset cleaning process is aimed to
be established, especially in scenarios where labeled data is
limited. Additionally, insights into the comparative strengths and
weaknesses of the few-shot learning approach against SVM, KNN,
and clustering models will be provided, offering valuable guidance
for practitioners in the fields of machine learning and data
preprocessing.
Through this study, a deep understanding of the potential of
few-shot learning in improving the dataset cleaning process is
aimed to be established, especially in scenarios where labeled data
is limited. Additionally, insights into the comparative strengths and
weaknesses of the few-shot learning approach against SVM, KNN,
and clustering models will be provided, offering valuable guidance
for practitioners in the fields of machine learning and data
preprocessing.
Paper Author Methodology
Metric-
Based
Learning for
Nearest-
Neighbor
Few-Shot
Image
Classification
Min Jun
Lee,
Jungmin
So
In this paper Uses triplet, cross-entropy,
and mixed losses during the embedding
network's training, we analyzed the
outcome for nearest-
neighbour classification. Using the
triplet loss has a major impact in the 1-
shot setting. The same loss showed the
best accuracy for the 5-shot setting in
the normalized configuration, and
similar accuracy was displayed for the
full setups. However, high-level
backbones like ResNet or DenseNet are
difficult to construct since the
recommended triplet loss training
model uses considerable GPU RAM.
Few-shot
Image
Recognition
for UAV
Sports
Cinematogra
phy
Emmanoui
l
Patsiouras,
Anastasios
Tefas,
Ioannis
Pitas
In this paper, we investigate the
behavior of such few-shot methods in
the context of drone vision
cinematography for sports event
filming, in order to recognize new
image classes by taking into
consideration the fact that this new
class we wish to identify is a subclass of
an already known class. More
specifically we use UAV footage to
recognize certain types of athletes,
belonging to a subset of an original
athlete class, utilizing only a handful of
recorded images of this athlete
subclass.
Few-shot
Fish Image
Generation
and
Classification
U.S. Gulf
Coast
The focus of this paper is on generating
realistic images for few-shot classes and
then using these images to enhance the
classification task. They conduct
experiments, increasing the number of
generated images to observe the
impact on classification accuracy.
Literature Review
• To efficiently train the large dataset using the 2D CNN model
with good accuracy
• To efficiently train the few shot images using the channel boost
CNN model with good accuracy
• To achieve the proper cleansing of the input image using the
hybrid model which includes CB-CNN and 2D CNN
Applications
• Biomedical Data Curation
• Object Detection in Autonomous Vehicles
• Social Media Analysis
• Recommendation Systems for E-commerce
• Anomaly Detection in Healthcare