duongngyn0510 / text-image-retrieval

Build end2end text-image retrieval app

Local service

Requirements

$ pip install git+https://github.com/openai/CLIP.git
$ pin install -r requirements.txt

Model

Fine-tune CLIP in image-retrieval task

• Input: Image or text query related to FASHION.

• Output: Top images with the highest similarity according to the cosine metrics.

Database

Using Pinecone vector database for fast retrieval result

• Vector database contains 85577 vector ids, those vectors are images embedding and their metadata.

Using Google Cloud Storage for storing image data

Local test

$ docker pull duong05102002/retrieval-local-service:v1.23
$ docker run -p 30000:30000 duong05102002/retrieval-local-service:v1.23

Run client.py for test the local api.

• Image query

$ python client.py --save_dir temp.html --image_query your_image_file

• Text query

$ python client.py --save_dir temp.html --text_query your_text_query

Note: Refresh the html page to display the images

• Top 8 products images similar with image query:

• Top 8 products images similar with text query: crop top

Response time (traces)

• Image query

• Text query