This is an agentic system to assist on basic image analysis such as feature extraction, size reduction and transformations (augmentation) using tool calling by the assistant's agent.
The resize tool just change the resolution in pixels of the image and save its reduced version, and it is very useful to test classification models in a low memory computing environment.
The transformation tool apply a transformation pipeline published in the article https://academic.oup.com/bioinformatics/article/35/21/4522/5466454 that has flips, rotations, zoom among the possible perturbations in the original image.
The feature extraction tool extracts 1d numerical vectors from each image and saves in a data table. The user may specify the method and the number of components in case the user wants to perform principal components analysis. The default method is vanilla, which just flatten the array, whereas the possible methods to choose are Truncated-SVD, PCA, KernelPCA using kernel linear and KernelPCA using kernel cosine.
- Create environment:
conda env create -f imgbot_env.yml - Activate env:
conda activate imgbot_env - Add the repo path in your PYTHONPATH env variable (in your ~/.bashc):
export PYTHONPATH="/path/to/imgbot_agent_repository:$PYTHONPATH" - Run the application using:
streamlit run index.py