Professional Documents
Culture Documents
Abstract
Summary of Requirements and Design ( Phase - 1)
Summary of Methodology / Approach (Phase - 1)
Design Description
Modules and Implementation Details
Project Demonstration and Walkthrough
Test Plan and Strategy
Results and Discussion
Lessons Learnt
Conclusion and Future Work
References
Abstract
• User Interface
• Audio to Text
• Image Generation
Set the context • Face Tuning
• Different Poses
• Custom Dataset
Summary of Requirements and Design
• Summary of Requirements (A bulleted list of the major requirements.
• Functional Requirements:
Process audio for face generation.
Generate 2D faces with variations.
• Software Requirements:
Operating System
Programming language and libraries/Frameworks
• Hardware Requirements:
Processor
GUI
Summary of Requirements and Design
Research Project
• Proposed Methodology
a) Model Architecture
Summary of Methodology / Approach
• Image Generation:
Objective: Generate facial images based on the input textual descriptions, focusing on realism and diversity.
Research Project
• The generator produces a new data point, such as an image, from an input random
noise vector.
• The generator is to provide data points that are comparable to the actual data
points.
• The discriminator attempts to differentiate between the genuine data points and the
created data points from the generator.
• The discriminator's objective is to correctly identify the real data points and tell
them apart from the phoney ones produced by the generator.
• As the generator attempts to trick the discriminator by producing better and more
realistic data points, the discriminator learns to better distinguish between the real
and generated data points. This adversarial training process for the GAN takes place.
Design Description
• The discriminator is updated in response to feedback from the generator, and both
the real and created data points are used to update the discriminator.
• The discriminator becomes better at telling the difference between actual and
produced data points, and the generator gets better at producing convincing data
points that deceive the discriminator.
• The generator should ultimately produce data points that are identical to the real
data points in order to produce a high-quality generative model.
Modules and Implementation Details
• Custom Dataset
Modules and Implementation Details
• Editing Script
Modules and Implementation Details
• Editing Script
Modules and Implementation Details
Generating
Poses script
Modules and Implementation Details
• MXNET Attribute
Classifier
Project Demonstration
Provide,
• Testing activities that are carried out along with
timeline.
• What are the test methods followed? and Why?
• (Example) – Functional Testing (Unit, Integration,
Note:
…) • Appropriate modifications can
• Non – Functional Testing (Performance, Security,
be done for Research Projects
…)
• What is the test environment? (Explain the role • Add as many slides as required
of each member in the team)
• Benefits of this approach & are there any
drawbacks?
• Test tools used? Automated test tools? Open-
source tools?
Results and Discussion
Are the results same as expected? Is it as per initial estimates planned? If there
is a deviation, give the reasons for the change.
Product based projects can explain how your product meets the requirements.
Clearly tie each test to the requirement (forward and backward traceability).
Schedule
Discussion of
how well the
schedule was
met.
You can add planned efforts (as per initial estimates) and actual
a table, efforts.
• Discuss on the lessons learned and what you could have done
differently knowing what you now know
Tao Xu, Pengchuan Zhang, Qiuyuan Huang, Han Zhang, Zhe Gan, Xiaolei
Huang, Xiaodong He “AttnGAN: Fine-Grained Text to Image
Generation with Attentional Generative Adversarial Networks” Published
in 2017
Ming Tao, Hao Tang, Fei Wu, Xiaoyuan Jing, Bing-Kun Bao, Changsheng
Xu “DF-GAN: A Simple and Effective Baseline for Text-to-Image
Synthesis” Published in 2022
Tingting Qiao, Jing Zhang*, Duanqing Xu*, Dacheng Tao, College of Computer
Science and Technology, Zhejiang University, China, School of Automation,
Hangzhou Dianzi University, China, UBTECH Sydney AI Centre, School of
Computer Science, FEIT, The University of Sydney, Australia: “MirrorGAN:
Learning Text-to-image Generation by Redescription”Published in 2021
Bowen Li, Xiaojuan Qi, Thomas Lukasiewicz, Philip H. S. Torr, University of
Oxford: “Controllable Text-to-Image Generation” Published in 2019