> ML_LITERATURE // CARION-2020-END-TO-END-OBJECT-DETECTION-WITH-TRANSFORMERS-DETR_v1.0
End-to-End Object Detection with Transformers (DETR)
Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko · European Conference on Computer Vision (ECCV) (2020)
seminal-architecture2020industry-standardthirdPartyReproduced
Principal Contribution
Eliminated hand-designed components like anchor generation and Non-Maximum Suppression (NMS) by framing detection as direct set prediction using a bipartite matching loss.
Operational Relevance
Serves as qualified reference for implementing task-object-detection in production systems.
Assumptions
- Underlying spatio-temporal continuity and domain distributional stability hold
Limitations
- Performance scaling and computational footprint depend on receptive field depth, sequence length, and resolution
Connected Algorithms, Architectures & Tools
Related Algorithms:
Related Architectures:
Implementing Libraries: