Object Detection Research Lab
A single-page interactive lab over the 48-paper corpus. Every claim is traced to a paper (P001…P048); every math object is executable in the Mathematics Lab; every section is one year of field state. Claims that are not in the corpus are flagged INSUFFICIENT EVIDENCE.
Corpus by year
The research ↔ application closed loop
Single-pass regression + multi-scale pyramids (2015-2020)
NMS-free, GPU/edge-class latency detection (2021-2024)
Cameras/robotics/drones/AV/open-vocab search (2017-2026)
Limited compute, quantization, occlusion, domain shift, latency
Efficient detectors, NMS removal, assignment design, domain adaptation
CSP backbones, anchor-free heads, re-param, NAS, dual-assignment, attention
Every deployed detector founds the next research problem: latency sources → efficiency, NMS failure → end-to-end, occlusion → robust matching, unseen classes → open-vocabulary.