Atten4Vis/ConditionalDETR
A transformer-based object detection model that achieves 6.7-10x faster training convergence than standard DETR on COCO.
Not currently ranked — collecting fresh signals.
star history
This repository implements Conditional DETR, an object detection model that modifies the transformer decoder’s cross-attention mechanism. The key innovation is a conditional spatial query that narrows each attention head to a specific image region, reducing dependence on high-quality content embeddings and easing training difficulty. The model integrates with Huggingface Transformers and achieves state-of-the-art convergence speeds on COCO 2017 validation.
Frequently asked
- What is Atten4Vis/ConditionalDETR?
- A transformer-based object detection model that achieves 6.7-10x faster training convergence than standard DETR on COCO.
- Is ConditionalDETR open source?
- Yes — Atten4Vis/ConditionalDETR is open source, released under the Apache-2.0 license.
- What language is ConditionalDETR written in?
- Atten4Vis/ConditionalDETR is primarily written in Python.
- How popular is ConditionalDETR?
- Atten4Vis/ConditionalDETR has 404 stars on GitHub.
- Where can I find ConditionalDETR?
- Atten4Vis/ConditionalDETR is on GitHub at https://github.com/Atten4Vis/ConditionalDETR.