Model description

Plain-DETR is an object detector that maintains a "plain" nature: using a single-scale feature map and global cross-attention calculations without specific locality constraints.
In contrast to previous leading DETR-based detectors that reintroduce architectural inductive biases of multi-scale and locality into the decoder.
By leveraging the Object365 dataset for pre-training, it achieved 63.9 mAP accuracy using a Swin-L backbone, which is highly competitive with state-of-the-art detectors which all heavily rely on multi-scale feature maps and region-based feature extraction
Published in the proceedings of ICCV 2023.

Open source status

• The model implementation is available
• The model weights are available

Provide useful links for the implementation

Official implementation: code
Paper: DETR Does Not Need Multi-Scale or Locality Design