Project page and code for ComCa are available
Published by Marco Garosi in Research · Monday 24 Mar 2025 · 1:00
Tags: attribute, detection, open, vocabulary, training, free, paper, code
Tags: attribute, detection, open, vocabulary, training, free, paper, code

Code for my conference paper ComCa is now openly available on GitHub, along with the project page.
The code has been developed with the latest libraries employed in deep learning, such as:
- PyTorch, as the main library to handle tensors, define and load models, and perform heavy computation;
- Transformers by HuggingFace, to load pre-trained 2D vision-language models such as CLIP and SigLIP;
- Faiss by Meta's Fundamental AI Research group to efficiently search in large collections of dense vectors. Specifically, the library is used to retrieve data from web-scale databases with millions of elements in an efficient way.
Code is available here.
There are no reviews yet.