Badger TeamX Video Retrieval

Developed from the AI Challenge 2022

Video retrieval interface based on AI.

Features

• Retrieve images based on english description, similar images, keywords, hybrids,...
• Context slider supports viewing frames around keyframes
• Select and load local or full data partition
• Access online youtube videos at any time or frame
• Queue, add new or delete selected frame at any position
• Export the queue as a file (*.csv) for the qualifying round
• Load, edit and submit final results via API
• Adjust output quantity, show real-time search results
• Manage search results based on multi-tabs

Installation and Running

• Requires at least 120GB of available storage
• Now available on Windows, MacOS and Linux
• Requirements: Python >= 3.8, <= 3.10.
• Recommended: Miniconda/Anaconda.

Download source code

2. Install git.
3. Download the badger-teamx-retrieval repository

git clone https://github.com/mrtrieuphong/badger-teamx-retrieval.git
cd badger-teamx-retrieval

Create environment

1. Use Conda [Recommended]

conda create -n badger python=3.8
conda activate badger-venv

2. Use virtualenv

pip install virtualenv
python3 -m venv badger-venv
# MacOS, Ubuntu
source badger-venv/bin/activate
# Windows
badger-venv/Scripts/activate

Install requirements

pip install -r requirements.txt

Download required

1. OpenAI CLIP model: ../CLIP/clip/bpe_simple_vocab_16e6.txt.gz
2. Images dataset: ../Images
3. Features: ../Features

Prepare Materials

1. Create thumbnails

python3 Tools/1_create_thumbnails.py

2. Create photo ids

python3 Tools/2_create_photo_ids.py

3. Create features

python3 Tools/3_create_features.py

4. Create mapping

python3 Tools/4_create_mapping.py