【YOLOv6 VS PP-YOLOE s（400 epoch） Update】官方性能实测更新 / Update of the Official Performance Comparison

[YixinKristy]

Update of the Test Result after Training 400 Epoch of PP-YOLOE s

Here is the latest comparison result:

Model	epoch	AP 0.5:0.95	AP 0.5	AP 0.75	AP small	AP medium	AP large	AR small	AR medium	AR large
PP-YOLOE s	300	43.0	59.6	47.2	26.0	47.4	58.7	45.1	70.6	81.4
PP-YOLOE s	400	43.4	60.0	47.5	25.7	47.8	59.2	43.9	70.8	81.9
YOLOv6 s	400	43.1	62.0	46.2	23.9	47.4	58.8	40.2	65.5	76.4

Feel free to raise any question or suggestion about the result~ PP-YOLOE always respect the original work of YOLO series and we'll keep working on providing more high-quality models.

For more reference: https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/configs/ppyoloe

The original issue with complete test result: https://github.com/meituan/YOLOv6/issues/140

性能对比测试数据更新（PP-YOLOE s训练400 epoch后）

2022.7.5日更新：

由于收到大家的反馈，我们训练针对PP-YOLOE s版本训练了400 epoch，以下是对比效果：

模型	epoch	AP 0.5:0.95	AP 0.5	AP 0.75	AP small	AP medium	AP large	AR small	AR medium	AR large
PP-YOLOE s	300	43.0	59.6	47.2	26.0	47.4	58.7	45.1	70.6	81.4
PP-YOLOE s	400	43.4	60.0	47.5	25.7	47.8	59.2	43.9	70.8	81.9
YOLOv6 s	400	43.1	62.0	46.2	23.9	47.4	58.8	40.2	65.5	76.4

大家有任何问题欢迎在评论区中讨论提出~

PP-YOLOE始终保持着向YOLO系列的致敬，希望大家能多多尝试，提出宝贵的建议！

如需获取完整的对比测试，请参考原始issue：https://github.com/meituan/YOLOv6/issues/140

[hutao568]

美团：这是来砸场子的吗Σ(⊙▽⊙"a

[qmcreeper]

美团：这是来砸场子的吗Σ(⊙▽⊙"a

再来个yolov7的对比就更有意思了 : P