Webcamdinov2
leverages the DINOv2 architecture to enhance video inferencing capabilities using a webcam, processing videos close to real-time. This implementation employs the MPEG-4 codec to display feature-extracted video sequences alongside the original sequences.
Repository: Webcamdinov2 on GitHub
xFormers
library.git clone https://github.com/1ssb/webcamdino
requirements2.txt
via pip:
pip install -r requirements2.txt
The inference pipeline integrates components adapted from Meta’s Facebook Research. Note that the current implementation does not achieve real-time performance due to the computationally intensive nature of the inferencing process. Efforts to reduce latency are ongoing, with potential future improvements through computational acceleration.
If you find this project useful, leave a star, or please cite it as:
@misc{bhattacharjee2023webcamdinov2,
author = {Bhattacharjee, Subhransu S.},
title = ,
year = {2023},
howpublished = {\url{https://github.com/1ssb/webcamdino}},
note = {Accessed: [Insert date here]}
}