search

PaddlePaddle / PaddleDetection

Object Detection toolkit based on PaddlePaddle. It supports object detection, instance segmentation, multiple object tracking and real-time multi-person keypoint detection.
Apache License 2.0
12.82k stars 2.89k forks source link

anchor_cluster.py生成的anchors训练的结果还没有自带的anchors精度高 #4201

Closed 534114658 closed 2 years ago

534114658 commented 3 years ago

PP-YOLO 使用PP-YOLO的ResNet18vd 原档anchors如下： anchors: [[10, 14], [23, 27], [37, 58], [81, 82], [135, 169], [344, 319]]

使用anchor_cluster.py生成anchors如下图 生成命令： python anchor_cluster.py -c annotations\train.yml -n 6 -s 416,256 -m v2 -i 1000 image anchors: [[30, 35], [37, 35], [36, 40], [47, 48], [53, 62], [59, 61]]

30张图片 都训练10分钟 自带的anchors训练效果很好，100%识别，识别框也很准确 用anchor_cluster.py生成的anchors训练 同样的参数训练有10%无法识别 还有些识别框不准确，如下图： image 正常标注如下： image

是哪里参数设置问题吗？

yghstill commented 3 years ago

@534114658 原本的anchor设置是在coco大规模数据上聚类得到的，鲁棒性很高，图片较少时不建议重新聚类了。或者要确认下anchors这个要在网络和reader中两处同时修改

paddle-bot-old[bot] commented 2 years ago

Since this issue has not been updated for more than three months, it will be closed, if it is not solved or there is a follow-up one, please reopen it at any time and we will continue to follow up. It is recommended to pull and try the latest code first. 由于该问题超过三个月未更新，将会被关闭，若问题未解决或有后续问题，请随时重新打开（建议先拉取最新代码进行尝试），我们会继续跟进。