TEKNOFEST 2024 E-Ticaret Hackathonu 3.'lük Derecesi Alan Proje
Python
3
14 commits
updated Oct 3, 2024
The goal of the Teknofest E-Commerce Hackathon, sponsored by Trendyol, is to enhance product photos uploaded by local sellers and generate more professional product descriptions using Generative AI.
[!NOTE] Special thanks to Trendyol for providing us with a productive and supportive working environment throughout the competition.
You can view the presentation here.
requirements.txt file using the following code:pip install -r requirements.txt
To get started, follow these steps:
app.py file using the following code:streamlit run app.py
pyngrok if you haven't already:pip install pyngrok
ngrok config add-authtoken "$YOUR_AUTHTOKEN"
ngrok http 8501
after setting it up a link will pop up indicated with Forwarding, click it to open the page.
[!NOTE] You have to Sign up for ngrok to able to run it. Obtain your authtoken from Your Autotoken page. Now you're ready to go and run the code 🎉
Our Streamlit application contains two tabs. The first tab is for background removal. To get the output, you simply need to:
| Before | After |
|---|---|
 |  |
The second tab provides an object removal feature. Here’s how it works:
| Before | After |
|---|---|
 |  |
In addition to the object and background removal features already present in our application, we've introduced an image upscaling function. After removing an object or background, the image is automatically upscaled to enhance its resolution. This ensures that both our model and the user can view even the finest details with clarity, delivering a more refined and high-quality result. This enhancement improves the overall user experience by providing sharper and more detailed images post-processing.
We've added a powerful text generation feature in both tabs of our application. After hitting the "Send" button, our models generate a detailed product description based on both the provided image and prompt. So to say, the Llama 3.2-11B-Vision model serves as the eyes of the Gemma-2-27B-IT model. We produce professional results with Gemma by incorporating the information extracted by the Llama Vision model.
As you can see in the first example, although we did not explicitly state that the product is "sugar-free," the Llama Vision model correctly identified it as such.
[!WARNING]
You may experience issues running this on less powerful hardware.
This project makes use of the following models:
google/gemma-2-27b-it:
meta-llama/Llama-3.2-11B-Vision:
ZhengPeng7/BiRefNet:
fal/AuraSR-v2:
diffusers/stable-diffusion-xl-1.0-inpainting-0.1:
[!NOTE] Please ensure compliance with each model's license when using or distributing this project.
TEKNOFEST 2024 E-Ticaret Hackathonu 3.'lük Derecesi Alan Proje
Python
3
14 commits
updated Oct 3, 2024
The goal of the Teknofest E-Commerce Hackathon, sponsored by Trendyol, is to enhance product photos uploaded by local sellers and generate more professional product descriptions using Generative AI.
[!NOTE] Special thanks to Trendyol for providing us with a productive and supportive working environment throughout the competition.
You can view the presentation here.
requirements.txt file using the following code:pip install -r requirements.txt
To get started, follow these steps:
app.py file using the following code:streamlit run app.py
pyngrok if you haven't already:pip install pyngrok
ngrok config add-authtoken "$YOUR_AUTHTOKEN"
ngrok http 8501
after setting it up a link will pop up indicated with Forwarding, click it to open the page.
[!NOTE] You have to Sign up for ngrok to able to run it. Obtain your authtoken from Your Autotoken page. Now you're ready to go and run the code 🎉
Our Streamlit application contains two tabs. The first tab is for background removal. To get the output, you simply need to:
| Before | After |
|---|---|
| |
The second tab provides an object removal feature. Here’s how it works:
| Before | After |
|---|---|
| |
In addition to the object and background removal features already present in our application, we've introduced an image upscaling function. After removing an object or background, the image is automatically upscaled to enhance its resolution. This ensures that both our model and the user can view even the finest details with clarity, delivering a more refined and high-quality result. This enhancement improves the overall user experience by providing sharper and more detailed images post-processing.
We've added a powerful text generation feature in both tabs of our application. After hitting the "Send" button, our models generate a detailed product description based on both the provided image and prompt. So to say, the Llama 3.2-11B-Vision model serves as the eyes of the Gemma-2-27B-IT model. We produce professional results with Gemma by incorporating the information extracted by the Llama Vision model.
As you can see in the first example, although we did not explicitly state that the product is "sugar-free," the Llama Vision model correctly identified it as such.
[!WARNING]
You may experience issues running this on less powerful hardware.
This project makes use of the following models:
google/gemma-2-27b-it:
meta-llama/Llama-3.2-11B-Vision:
ZhengPeng7/BiRefNet:
fal/AuraSR-v2:
diffusers/stable-diffusion-xl-1.0-inpainting-0.1:
[!NOTE] Please ensure compliance with each model's license when using or distributing this project.