# 视觉与 YOLO 文献脉络

> 本页由仓库源文件生成。请不要直接编辑本页；修改下方 source 文件后运行 `vp run docs:generate`。

> 中文支持说明：本页是中文站点镜像，保留论文标题、证据状态和来源字段的原始表述，避免未经核实的翻译改变证据边界。

> 来源: `literature/03-yolo-history-and-paradigms.md`

## YOLO History And Paradigms

Status: `needs confirmation`

## Review Questions

* What makes one-stage detection suitable for real-time candidate generation?
* How should `bbox`, `class`, and `confidence` be transformed into BCI command candidates?
* How should candidates be filtered, tracked, frozen, and confirmed?
* What failure modes matter for BCI selection and robot grasping?
* What additional method is needed for grasp-pose estimation beyond object detection?

## Evidence To Extract

| Evidence Need | Candidate IDs | Status |
| --- | --- | --- |
| YOLO real-time detection baseline | YOLO-001 | text extracted |
| YOLO evolution | YOLO-002, YOLO-003 | text extracted |
| two-stage comparator | YOLO-004 | text extracted |
| dataset/context evidence | YOLO-005 | text extracted |
| dense detector comparator | YOLO-007 | text extracted |
| modern YOLO variants | YOLO-009, YOLO-010 | text extracted |
| RGB-D / hand-eye / grasp pose | GRASP-001, GRASP-003 to GRASP-007 | text extracted |