丢弃Transformer，FCN也可以实现E2E检测

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我们基于FCOS，首次在dense prediction上利用全卷积结构做到E2E，即无NMS后处理。我们首先分析了常见的dense prediction方法（如RetinaNet、FCOS、ATSS等），并且认为one-to-many的label assignment是依赖NMS的关键。受到DETR的启发，我们设计了一种prediction-aware one-to-one assignment方法。此外，我们还提出了3D Max Filtering以增强feature在local区域的表征能力，并提出用one-to-many auxiliary loss加速收敛。我们的方法基本不修改模型结构，不需要更长的训练时间，可以基于现有dense prediction方法平滑过渡。我们的方法在无NMS的情况下，在COCO数据集上达到了与有NMS的FCOS相当的性能；在代表了密集场景的CrowdHuman数据集上，我们的方法的recall超越了依赖NMS方法的理论上限。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"整体方法流程如下图所示："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/67\/c1\/67436c182b11e6bf5afa97f183d446c1.png","alt":null,"title":"","style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":"","fromPaste":false,"pastePass":false}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"One-to-many vs. one-to-one"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}}]}