Stylist Ai : Revolutionizing Fashion with AI Outfit Recommendation, Outfit Generation, Virtual Try-On, and Fashion Chatbot! 👗🤖

👗✨ Experience the future of fashion with Stylist AI, where innovation meets style. Our cutting-edge platform revolutionizes your wardrobe by offering personalized outfit recommendations and generating unique ensembles tailored just for you. Explore our virtual try-on feature to visualize your perfect look effortlessly. Engage in fashion-forward conversations and get expert styling advice with our AI-powered chatbot. Elevate your fashion game with Stylist AI and discover a new era of effortless elegance. 🌟🤖

Demonstration of the Project

https://github.com/dhaan-ish/intelOneApiHackathon/assets/119067139/d0834784-0e5a-4795-b0e0-49b617a9e05d

Click here to watch the demo video

Stylist Ai : Five Models

1. Text-To-Outfit-Generator 📝👗
2. Human-Identification 👤🚀
3. Outfit-Recommendation 🛍️💡
4. Virtual-Try-On 🌐👀
5. Chat-Bot 💬🤖

<a name="Text-To-Outfit-Generator"></a>

Text-To-Outfit-Generator 📝👗

This code snippet demonstrates the utilization of a Text-Outfit-Generator model, leveraging a pretrained model from Hugging Face. The model is part of the DiffusionPipeline package, facilitating the generation of outfit descriptions based on textual prompts. By loading the model onto the available device, the script efficiently processes the input prompt and produces an image of the described outfit. Notably, running this code in Google Colab exceeded 15 minutes, but leveraging Intel's CPU or XPU ensures completion in less than a minute. ⚙️👗🚀

Hugging Face URL : Fashion-Product-Generator

Prompt : Flowers in green shirt with white colored button.

Folder Link : Click Here

<a name="Human-Detection"></a>

Human-Identification 👤🚀

This code implements an image classification model using TensorFlow and Keras. The model is designed to classify images into two classes: "human" and "non-human." The neural network architecture consists of convolutional (Conv2D) and pooling (MaxPooling2D) layers, followed by fully connected (Dense) layers. The model is compiled using the binary crossentropy loss function and the Adam optimizer. The Intel OneDNN helped in reducing the time for training, and the optimized TensorFlow for Intel Hardwares helped us in reducing the time for training. 🌐🖼️🤖💪

Folder Link : Click Here

<a name="Outfit-Recommendation"></a>

Outfit-Recommendation 🛍️💡

This code snippet demonstrates the implementation of an outfit-recommendation model using Semantic Image Search techniques. Leveraging the Langchain library, it combines BM25 retrieval with dense vector retrieval using CLIP embeddings. The outfit recommendation process involves encoding sparse and dense vectors from fashion metadata and images, persisting them in a vector database, and utilizing retrievers to search and display relevant outfit suggestions. By integrating both textual and visual cues, this model enhances the accuracy and diversity of outfit recommendations based on the color and ocassion given by user. catering to a wide range of fashion preferences. 🧥📊👠👗🎨

<div style="display: flex;"> <img src="http://assets.myntassets.com/v1/images/style/properties/27f41de6e52f15c6e11b8fc4ae98e889_images.jpg" width="200" alt="Image 1"> <img src="http://assets.myntassets.com/v1/images/style/properties/Reid---Taylor-Men-White-Shirt_7868945b6d3a95140b08b1719b0092f1_images.jpg" width="200" alt="Image 2"> <img src="http://assets.myntassets.com/v1/images/style/properties/ea2bbdc84a4512d5987e74c557e2b141_images.jpg" width="200" alt="Image 3"> <img src="http://assets.myntassets.com/v1/images/style/properties/Arrow-Men-White-Striped-Shirt_e9c86c6203c4b2ac6d9739069ebc71e7_images.jpg" width="200" alt="Image 4"> </div>

Selected Ocassion : Wedding Selected Color : White

Folder Link : Click Here

<a name="Virtual-Try-On"></a>

Virtual-Try-On 🌐👀

This code implements a Virtual-Try-On (VTON) model that allows users to try on clothes virtually using an input image. The model consists of several steps, including clothing segmentation, pose estimation, and image composition.

1. Clothing Segmentation: The get_cloth_mask.py script utilizes a pre-trained model to segment clothing from the input image, generating a mask that isolates the clothing item. 🕵️‍♀️👕

2. Pose Estimation: The posenet.py script estimates the pose of the person in the input image using a PoseNet model. It detects key body keypoints, necessary for aligning the clothing item with the person's body. 🤸‍♂️📏

3. Image Processing: The main process in main.py combines the segmented clothing mask and the pose estimation results to compose a final image of the person wearing the virtual clothes. It involves resizing the input image, generating semantic segmentation, removing the background, and adding the clothing item onto the person in the image. 🔄📸

4. Final Output: The model produces a final image (finalimg.png) that showcases the person wearing the virtual clothes. The user can choose to retain the original background or remove it from the final composition. 🎉👚📷

Folder Link : Click Here

<a name="Chat-Bot"></a>

Chat-Bot 💬🤖

This code configures a chatbot using the Mistral 7B model, fine-tuned with a custom dataset. It employs various libraries and tools for setting up the chatbot environment, transformers, LLAMA Index, and Langchain embeddings. The process involves installing necessary dependencies, downloading the custom dataset (in PDF format), loading the dataset into LLAMA Index, configuring the Mistral 7B model for language generation, and setting up the Langchain embeddings for semantic understanding. Finally, it initializes the chatbot's query engine and demonstrates a sample query for testing purposes. 🤖📊🔧🗣️

Prompt : <br/> <br/> Output : <br/> <br/> Prompt : <br/> <br/> Output : <br/>

Folder Link : Click Here

Flow Diagram 🔄📊

The flow diagram illustrates the sequential steps and interactions within our system. Each stage in the process contributes to the overall functionality, ensuring a smooth and efficient workflow. Let's delve into the key components:

1. User Input 🤖🗣️:

   • Users initiate the process by providing input, whether through text prompts or selects ocassion.

2. Text-To-Outfit-Generator 📝👗:

   • The Text-To-Outfit-Generator module interprets textual prompts and generates corresponding outfit descriptions.

3. Human Detection 👤🚀:

   • The Human Detection component identifies and locates individuals within images. This step is crucial for subsequent processes, ensuring accurate and perfect virtual try on.

4. Outfit Recommendation 🛍️💡:

   • Based on selection of ocassion and color, the Outfit Recommendation module suggests personalized clothing ensembles. It considers factors such as style preferences, occasion, and user demographics.

5. Virtual Try-On 🌐👀:

   • Users have the opportunity to virtually try on suggested outfits. The Virtual Try-On feature utilizes image processing, clothing segmentation, and pose estimation to showcase how the recommended outfits would look on the user.

6. Fashion Chatbot 💬🤖:

   • Engaging with users in natural language, the Fashion Chatbot provides additional assistance, answers queries, and offers styling advice. It enhances the overall user experience through interactive and dynamic conversations.

Built With 🛠️

1. Frontend - React: Our frontend user interface was developed using React, a popular JavaScript library for building user interfaces. React's component-based architecture allowed us to create modular and reusable UI components, facilitating the development of a responsive and interactive user experience. 💻🌐

2. Backend - Flask: The backend of our application was built using Flask, a lightweight and flexible web framework for Python. Flask prov