Feature:The camera can detect whether pets are eating food that is harmful to their health.

Reason:Owners may sometimes overlook their pets or be unable to care for them while working outside. A camera can help monitor all of the pet's activities, allowing owners to keep an eye on their pets even when they are away. This can prevent pets from accidentally eating food that is harmful to their health, ensuring they stay safe and healthy while their owners are not home.

Existing Project 2

Feature:Message transmission

Reason:When the camera detects food that pets should not eat, the project will send a message to the owner, alerting them that their pet is close to something harmful. This real-time notification helps owners take immediate action to prevent their pets from consuming dangerous items. By keeping owners informed about their pets' activities, the system promotes better pet care and safety, ensuring that pets stay healthy and avoid potential hazards.

Existing Project 3

Feature:buzzer

Reason:As mentioned above, when the camera detects food that pets should not eat, the project will send a message to the owner and also trigger an alarm sound to warn the pet to stay away from the harmful food. Additionally, the owner can record their own custom sound to serve as a warning alarm when the pet approaches the food. This dual alert system not only keeps the owner informed but also actively discourages the pet from getting too close to dangerous items, enhancing overall pet safety.

AI遠端應診服務

Feature:

Instant condition analysis and diagnosis

Reason: Improve access to healthcare

In remote areas or places with limited medical resources, the number of professional doctors is limited. AI remote consultation service allows patients to make diagnosis and treatment recommendations through intelligent systems or professional doctors without being restricted by geographical location, allowing more people to enjoy timely medical services.

Existing Project 2

AI醫療機械人

Feature:Quickly and accurately analyze medical images

Recognize early signs of disease

Cross-reference large amounts of medical data

Provide initial diagnostic advice

Reason:

Enhanced Precision and Accuracy: AI-powered robots can perform surgical procedures with a level of precision that surpasses human capabilities. This can lead to fewer complications, reduced blood loss, and quicker recovery times for patients.

Existing Project 3

AI-driven packaging design consumer research (PLUG)

Feature:Consumer Insights and Preferences:

• Sentiment Analysis: AI analyzes consumer reviews, social media, and feedback to gauge sentiment towards packaging designs. This helps designers understand what consumers like or dislike.

• Trend Analysis: AI can identify emerging trends in consumer preferences and behaviors, allowing brands to stay ahead of the curve.

• Reason:Consumer-Centric Design:

• Understanding Preferences: AI-driven consumer research allows brands to gather and analyze large volumes of consumer data to understand preferences, behaviors, and trends. This insight ensures that packaging designs align with what consumers actually want and need.

The second part of the image describes the second function of the workflow. This involves performing ArtiFade fine-tuning by minimizing the reconstruction loss between the clean input image and the generated image with its corresponding blemished embedding. The researchers evaluate the performance of their method using a fine-tuned U-Net model with a new type of blemished embedding.

The image includes several example images demonstrating the input, the ArtiFade output, and the technical inversion results. It also shows a diagram of the training dataset and reconstruction loss evaluation process.

Overall, the image presents a novel technique for improving image clarity by leveraging a diffusion model to reduce noise, while also fine-tuning the model to minimize reconstruction loss between the original and generated images. This approach aims to enhance the visual fidelity of the output while preserving the key details of the original object. The researchers indicate they will apply this method to a new type of blemished embedding if the initial results are successful.

Feature: fine tuned diffusion model (微调扩散模型)

Reason:The input image appears to be quite noisy, with various distractions and artifacts that obscure the clear view of the original object, which seems to be a car. By applying a fine-tuned diffusion model, the goal is to selectively remove or reduce this unwanted noise from the input image, so that the output image can more clearly display the original object of interest - in this case, the car.

Diffusion models are a powerful tool for image denoising and enhancement because they can learn to capture the underlying structure and patterns in the data, allowing them to distinguish between the signal (the desired object) and the noise (unwanted artifacts and distractions). By fine-tuning the diffusion model on relevant datasets, it can become specialized in effectively removing the type of noise present in the input image, without compromising the integrity of the original object.

The end result, as shown in the output images, is a clearer, more focused representation of the car, with the unwanted noise and distractions significantly reduced or removed. This can be beneficial for a variety of applications, such as object detection, image analysis, or visual quality improvement.(provide for picture to use)

Existing Project 2

Feature:ArtiFade

Reason:The reasoning behind this approach can be explained as follows:

1.Reconstruction loss minimization: By minimizing the reconstruction loss between the clean input image and the generated output image, the fine-tuning process ensures that the output image maintains a close resemblance to the original, preserving the key details and features of the input.

2.Blemished embedding handling: The workflow accounts for the presence of blemished embedding in the generated image. This suggests that the input image may contain unwanted artifacts or distortions that need to be addressed. By incorporating the blemished embedding into the fine-tuning process, the method aims to mitigate the impact of these imperfections on the output image.

3.U-Net model fine-tuning: The use of a fine-tuned U-Net model, which is a widely adopted convolutional neural network architecture for image-to-image tasks, indicates the researchers' efforts to leverage advanced deep learning techniques to enhance the image reconstruction and denoising capabilities.

4.Iterative improvement: The mention of "if the first one successful, we will do this too" suggests a potential iterative approach, where the fine-tuning process can be further refined and improved based on the initial results, leading to increasingly better image quality and artifact removal.

Existing Project 3

Feature:AI-powered photo filtering system for kindergartens 幼兒園人工智慧照片過濾系統

Reason:

The second function aims to use a fine-tuned ArtiFade workflow to enhance blurry or out-of-focus images. By minimizing the reconstruction loss between the clean input and the generated output, the method can effectively reduce unwanted artifacts and improve the clarity of the final image, restoring the original details and features.

Project background

We are going to do a essay correction bot, the use of an essay correction bot can be beneficial in various ways, both for students and writers looking to improve the quality of their writing.Here are some function of it.First Grammar and Spelling Correction, the bot can help identify and correct grammar and spelling mistakes in essays, ensuring that the writing is error-free,human would not never right even a profession from university makes wrong,but bot won’t ,after learning,they would not make errors again.Nowaday,there are lack of resources which can teach you,only a few teacher in a school facing huge amount of student.“Transforming Education in Africa,” African Union, 2021. More specifically, according to the report, approximately 42 million children of primary and secondary school age are not enrolled in school and  estimated 15 million new teachers are needed by 2030.

If we are going to use this product,the resources of the bot need to learn is a lot.The skills of English grammar and Chinese grammar is totally different.And each of them have a long history of grammar growth.

(250 words)

Existing Project 1

Feature:

User tell their writing idea to the AI chatbot,  the chatbot may tell them how to write the good passage. 

Reason:

The chatbot will teach user the writing technique, like the grammar ,spelling,segmentation,feelings and the adjective to emphasize the feeling.

Existing Project 2

Feature: 

The chatbot can analyze the photo which user updated.

Reason:

Analyze the photo which user uploaded ,it can convenient user to simpifly the message and give suggestion to user. 

Existing Project 3

Feature:

translation

Reason:

User can send chinese(traditional) or chinese(simplified) to the bot ,the bot will translate it to english,it convenient user to write their passage and done the passage with some excel words 

After discussion, your group decided to refer to the features from…

Nowadays, with the advancement of artificial intelligence, there are various applications consists of the implementations of A.I . For example, ChatGPT is one of common application that is built with a large language model and trained with deep learning network.

Among all the current applications, image generation is one of the most useful one. It is because it can be highly applied in different fields such as retails, design, educations, etc. With a convenient tool, people can use A.I to generate their desired images.

In this way, our project is to build a similar application. The application aims to observe the key features of a specific input image and output an image based on the feature obtained, not to mention removing the noise in the original image.

Core Feature 1 -- removal of noise

It is worth noting that people need to search for some image online for various purpose such as research, design or even writing this proposal. However, images found online have high resolution or clean, say the image can be blurred.

To tackle this issue, our project can utilize the deep learning model to reconstruct the image by removing the noise, so the output image can be clean for usage. As the model is fine-tuned and given a dataset consists of clean images and noised images, it can have a highly accuracy on reconstructing the image while reducing features loss.

Core Feature 2 -- generation of image based on key feature

Instead of noise reduction, the project also can generate a new image given the original image by capturing the important features. Although people can freely use any images online, it may lead to infringement of copyright.

To prevent such problem, the model can read and understand the input prompts, then capture the significant features and reconstructing a brand-new image. Even though the image is reconstructed, the ley features are kept so that the user can use the image without violating the law.

Core Feature 3 – accept any custom input prompts

Different users will have various desired input, so our model can accept any input. The model is trained with deep learning network, meaning that it is “smart” enough to distinguish the feature among a wide range of images type and theme.

In this way, users from different ages or different fields can get their expected output image