GABAR (Gender & Age Based Actor Retrieval) is a computer vision project realized on the occasion of the final exam of the Visual Information Processing and Management course held at the University of Milano-Bicocca.
Given an image with one or more faces, we aim to detect them and estimate their gender and age. With these information we will propose an actor that resembles the selected person in the image. We have made a telegram bot as an interface to allow user interaction.
A detailed explanation of the entire project is available at visual-processing-project/GABAR.pdf or here
-
model/: Containing all the models used by the telegram bot. The pretrained models can be downloaded here -
src/: Project codemodels/detection/retrieval/telegram_bot/
-
notebooks/: Utility notebookmetadata_conversion_matlab2pandas.ipynb: Convert metadata from matlab format to pandas format.Gender_Age_Prediction_Using_VGGFace.ipynb: Performs the actual training of the finetuned VGG-Face model
-
dataset: Containing all the datasets used for training the models. Some of the datasets are currently not inside the folder because of Github storage limitations. You can find them in the following links:- FDDB
- UTKFaces (already inside the folder)
- Caltech256
- The dataset used for similar actor retrieval was scraped by us and can be found entirely in the
dataset/Retrievalfolder
pip install -r requirements.txt
- Generate a new telegram bot using BotFather
- Insert the bot token inside
src/config.pyin the variable TELEGRAM_BOT_TOKEN - Download from here the models and add them inside
visual-processing-project/model - Run
src/telegram_bot/bot.py