README.md

OneStopVision

Unleash the power of computer vision in your projects! OneStopVision is your one-stop shop for pre-trained algorithms, offering a comprehensive toolkit for facial analysis, object detection, and depth estimation. Dive into tasks like face recognition, landmark extraction, and head pose estimation – all readily available and accompanied by a user-friendly README for smooth installation and integration.

Features at a Glance

Feature	Description
Face Detection	OneStopVision detects human faces, returning bounding boxes and key landmarks in a convenient JSON format.
Face Recognition	OneStopVision calculates inter-face cosine similarities for recognition, extracting unique identity features in a single step.
Facial Attribute Analysis	OneStopVision analyzes age, gender, and emotion with probabilistic insights, unlocking a deeper understanding of your visual content.
Face Parsing	OneStopVision isolates facial regions with intelligent parsing, generating detailed masks for unparalleled control over your visual data.
Landmark Extraction	OneStopVision extracts 68 key landmarks and delivers them in convenient JSON format.
Head Pose Estimation	OneStopVision estimates head pose in yaw, pitch, and roll, even visualizing it directly on the image for an intuitive understanding of facial orientation.
ControlNet Operations	OneStopVision empowers you with M-LSD, HED, OpenPose, depth estimation, and semantic segmentation, all visualized for seamless integration and groundbreaking visual analysis.

• If you want to see how the features work, visit README_VIS.md.
• You can check what will be added as a feature in the next steps from the TODO.md file.

Install

conda create --name onestop python==3.9.18 -y
conda activate onestop

# CUDA 11.8
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118
pip install -r requirements.txt

bash downloadModels.sh

Run

steamlit run app.py

Acknowledgement

• Face Detection: The structure used for face detection is taken from the DSFD-Pytorch-Inference repository.
• Face Recognition: The model used for face recognition was taken from the insightface repository. Arcface R100 backbone was used to compare the identities.
• Facial Attribute Analysis: MiVOLO repository was used for Gender and age estimation. ResidualMaskingNetwork repository was used for emotion recognition.
• Face Parsing: The face-parsing.PyTorch repository was used for face parsing.
• Landmark Extraction: The face-alignment repository was used for 68 landmark extraction.
• Head Pose Estimation: The 6DRepNet repository was used for head pose estimation.
• ControlNet Operations: The controlnet-aux library, lllyasviel/ControlNet models were used for ControlNet operations.