Skip to content

AMAZON DEV HACKATHON 10TH RANK Using Visual AI technology to make amazon ecommerce more seamless, personalized, engaging more easier and flexible , allowing customer to search any buyable product by just capturing picture from mobile or directly choosing the product from video frame of amazon prime videos.

Notifications You must be signed in to change notification settings

Gaurav05082002/Visual_Image_Search

Repository files navigation

Visual_Image_Search | SHORT_DEMO_VIDEO_LINK |

Testing

Run Main_Searching_Model_Script.ipynb in a notebook environment for seeing completely error-free working demo Please refer to the Setting up the environment section for detailed setup instructions.

Currently, our model is trained mostly on women's wear products, so please use traditional wear images like saree, kurta, etc., as input.

Setting up the environment

Note: The model files are large (1.29 GB, 1.30 GB), so please use Google Drive to avoid errors.

  1. Download the Models folder from here.
  2. Upload it to your Google Drive.
  3. Mount your Google Drive in your notebook environment.
  4. Modify the paths for prediction_by_knn_model.pkl and product_maped_to_features in your code according to your mounted Google Drive.
  5. Install the required package: pip install ultralytics

Note: When running the cell for the first time, it may take longer. Please wait for some time.

Solution

Part 1: Search by Image of Single Product

Code to retrieve the best similar product link when uploading an image of a single product. The model performs image scan search and returns a link to a similar product from our demo data on which the model is trained. In the future, the model can be trained on millions of product images to return exact matches.

Part 2: Shopping from Prime Videos

When watching Prime Video, clicking the "Shop Now" button sends the video's time frame/screenshot to this model ( written as part 2 in Main_Searching_Model_Script.ipynb file. The model give cutouts of products in video frame then user select the product cutout which he wants to search, once selected it returns a link to a similar product from our demo data on which the model is trained. In the future, the model can be trained on millions of product images to return exact matches.

Data Used

Scraped Data

We scraped data from Amazon products using amazon_scrapper.ipynb. The model utilized around 4000+ products from the following categories, each file containing approximately 100 products summed up in all_combined.csv:

  • women professional dress
  • women party wear
  • women clothing
  • watches
  • traditional wear for women
  • stylish tops for women
  • saree
  • salwar suit
  • lehenga
  • ghagra
  • frock for women
  • formal dress for women
  • fashionable dresses for women
  • embroidery kurta
  • ethnic wear for women
  • embroidery kura
  • crop tops
  • heavy sarees
  • printed kurtas
  • red saree
  • bicycles
  • bottles
  • cartoys
  • comforters
  • fridges
  • jackets
  • tables
  • televisions
  • washing machines

All these data files can be found in the Scraped Data folder.

About

AMAZON DEV HACKATHON 10TH RANK Using Visual AI technology to make amazon ecommerce more seamless, personalized, engaging more easier and flexible , allowing customer to search any buyable product by just capturing picture from mobile or directly choosing the product from video frame of amazon prime videos.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published