DETR

DETR (DEtection TRansformer) is a model that treats object detection as a direct set prediction problem, consisting of a set-based global loss that forces unique predictions via bipartite matching and a transformer encoder-decoder architecture. Using a small set of learned object queries, it reasons about the relations between objects and the global image context to output the final predictions in parallel. It states that with a ResNet-50 backbone it achieves the same COCO 42 AP as Faster R-CNN with half the computation. The repository provides PyTorch training code, pretrained models, a Colab notebook, and a Detectron2 wrapper.

• End-to-end object detection based on a transformer encoder-decoder
• Set loss based on bipartite matching
• 42 AP on COCO (ResNet-50, half the compute of Faster R-CNN)
• Inference in 50 lines of PyTorch
• Pretrained DETR and DETR-DC5 models (Model Zoo)
• Detectron2 wrapper provided

Open source (free, weights/code released)

Released free under an open-source license with no extra cost for self-hosting (usage-based pricing when offered as a cloud API)

Estimate based on AI knowledge - please verify on the official site

• One of 22 AI tools in the Multimodal & Perception field.
• Connected to 7 other tools — #1 in its field.