ZIMA, S. Creating a Python-based Automated System for Recognizing Emotions from Facial Expressions. [online]. Brno: Vysoké učení technické v Brně. Fakulta informačních technologií. 2024.
| Kritérium | Známka | Body | Slovní hodnocení |
|---|---|---|---|
| Informace k zadání | The bachelor's thesis, titled "Creating a Python-based Automated System for Recognizing Emotions from Facial Expressions," focuses on developing a facial expression recognition (FER) system using deep learning. The thesis was quite challenging due to the complexity of recognizing subtle facial expressions and the need for an efficient model that could be deployed on devices with limited computational resources. The project builds on existing methods in the field but introduces a novel approach using a customized SqueezeNet architecture enhanced with dropout and batch normalization layers to achieve higher accuracy. The student achieved an accuracy of 66.37% on the FER-2013 dataset, which is satisfactory given the constraints and the nature of the task. | ||
| Práce s literaturou | The student demonstrated considerable initiative in obtaining and utilizing study materials. They extensively reviewed literature on human emotions, facial expressions, and existing FER methods. This thorough background research informed the design and implementation of the FER system, ensuring that the chosen methodologies and techniques were grounded in current scientific understanding and technological capabilities. | ||
| Aktivita během řešení, konzultace, komunikace | The student was highly active throughout the project, consistently meeting agreed deadlines and regularly consulting with the supervisor. They prepared thoroughly for consultations, presenting their progress and discussing challenges. The student showed dedication to refining their model, experimenting with various neural network architectures, and addressing overfitting issues through data augmentation and model regularization. He also discussed issues related to dropout layers. | ||
| Aktivita při dokončování | The student was proactive and consistent in their approach. Despite the inability to meet physically, regular online meetings ensured continuous discussion and feedback on the project's progress. The student was responsive and incorporated suggested improvements in a timely manner, demonstrating a good grasp of the required changes and their implementation. The deadlines were met consistently, and the student was well-prepared for each consultation, showing a deep understanding of the project's requirements and progress. | ||
| Publikační činnost, ocenění | There is no mention of specific publishing activity related to this work, such as journal papers. However, the thesis contributes valuable insights into developing efficient FER systems and may serve as a foundation for future publications or software releases in this domain. The student is planning to publish the results in a conference based on his thesis. |
Overall, the work presented in the thesis meets the requirements for a Bachelor thesis. Though the main shortcoming is a shallow literature review, the student did study the problem at hand, proposed a model for the problem, implemented it, and provided the results with sufficient number of quality metrics. Comparison and visualization of the results could have been better but a shallow literature review limited these aspects of the thesis.
| Kritérium | Známka | Body | Slovní hodnocení |
|---|---|---|---|
| Náročnost zadání | It is a difficult project because it requires not only knowledge in image processing but also needs to incorporate psychological concepts of human emotions. The challenges in recognizing various different emotions from facial features, irrespective of age and gender, makes it a difficult project. | ||
| Rozsah splnění požadavků zadání | The main shortcoming of the thesis is literature review. Since the student did not study the relevant articles, it was not possible for the student to provide comparison of the results of the proposed method. Overall, it is satisfactory because the student did study the problem, conducted a shallow literature review, proposed a model for the problem, implemented it, and provided the results with good number of quality metrics. | ||
| Rozsah technické zprávy | The thesis meets the minimum 40 standard pages requirement. Though it meets the requirement, the student could have easily expanded it by including details of literature reviewed which is not adequately discussed in the thesis. This information is lacking in this thesis. The student might have studied about it on his own but it does not reflect in the thesis. | ||
| Prezentační úroveň technické zprávy | 80 | Overall the structure of thesis, in terms of chapters organization, is satisfactory. Chapter 1 provides introduction, followed by information on human emotions, relation with facial expressions and the corresponding literature review in chapters 2 and 3. Chapters 4 and 5 provides details of the proposed method and the corresponding implementation of the proposed method. Results are discussed in chapter 6 followed by conclusion and future work in chapters 7 and 8. Separate chapter 8 was not required and both chapters 7 and 8 could have been combined. This is a minor observation. However, the major observation is the missing flow between the sections in the chapters and this makes this thesis harder to read. Also, many sections are one paragraph section. A section should have few paragraphs to discuss the topic of the section. | |
| Formální úprava technické zprávy | 80 | In terms of language, the thesis is readable and language of the thesis is satisfactory. There are minor typos and grammatical mistakes. However, they are few and in general the thesis is fine. The major observation is the linguistic aspect in terms of writing the thesis. It is more written like a layman rather than a proper thesis. The usage of “I” and using expressions like “I should try” etc should have been avoided. | |
| Práce s literaturou | 70 | This is the weakest part of the thesis. Though the student has used chapters 2 and 3 to provide the relevant details, however, the details of the existing state-of-the-art work in the field of FER is missing in the thesis. For example, for deep learning, only CNN is mentioned. Overall, student has not mentioned the gaps in current research and has not done satisfactory literature review. | |
| Realizační výstup | 90 | The images are converted from grayscale to colour which is not right approach for this problem. The student has used SqueezeNet based on CNN to solve the problem. SqueezeNet was released in 2016 and hence its an old model. Though good for learning, a more recent model should have been tried. Student is proposing some modifications to SqueezeNet which is good but without literature review, it is not possible to see whether they can be considered a contribution. Overall, the proposed model, its training, validation and testing are satisfactory for a BS thesis. | |
| Využitelnost výsledků | This thesis involves application of a deep learning method for FER (Face Emotion Recognition). The student did not provide comparison with state-of-the-art in FER. Hence, it is not possible to assess the possibility of using the results in practice, though it is possible in general for such a thesis. |
eVSKP id 156771