hulinlang

Results 4 comments of hulinlang

P,R一直为0,也预测不出框。

> [#386 (comment)](https://github.com/hukaixuan19970627/yolov5_obb/issues/386#issuecomment-1186461244) 按照这个回答重新发一次哦,提供的信息不大详细。 result ![image](https://user-images.githubusercontent.com/91964911/181243808-381a3a3b-22f7-4eb8-a157-48dbe7b4d97d.png) opt ![image](https://user-images.githubusercontent.com/91964911/181243844-269af03f-cfe4-4c9b-975b-0df9503f5a6b.png) ![image](https://user-images.githubusercontent.com/91964911/181243883-2bf949ab-cf16-410c-b2e5-6dafac1ffae9.png) hyp ![image](https://user-images.githubusercontent.com/91964911/181243924-26fd9e92-0cec-4b2d-9764-95eefb4383a4.png) ![image](https://user-images.githubusercontent.com/91964911/181243948-4f4cfa9c-2754-413d-ac1d-ad97a87ab97b.png) train_batch ![image](https://user-images.githubusercontent.com/91964911/181243991-576331d8-daae-4288-8c04-a4a0a5d39dee.png) ![image](https://user-images.githubusercontent.com/91964911/181244012-3eb2fd67-eae9-4f04-ad61-76f36d5cd6e2.png) ![image](https://user-images.githubusercontent.com/91964911/181244060-cc9947b8-9994-474a-993d-ca6026829351.png) val_batch_labels ![image](https://user-images.githubusercontent.com/91964911/181244136-f749bcb6-763d-4d3f-82e4-92ae414f5c33.png) val_batch_pre ![image](https://user-images.githubusercontent.com/91964911/181244176-9d8b5ebd-8e42-4de7-80b7-3bd5df604da9.png) 没有labels的那两个文件。 关于数据集 标注如下 ![image](https://user-images.githubusercontent.com/91964911/181244349-73484630-cda0-4869-bea0-8f4dcee5aa56.png) ![image](https://user-images.githubusercontent.com/91964911/181244400-92ed9516-009f-4325-9880-73624afa17f5.png) 其中有一些带负数的标注,运行时报警告舍弃了。 图像如下 ![image](https://user-images.githubusercontent.com/91964911/181244575-44851665-025c-450a-96ea-7380246b467d.png) 全为1024*1024的SAR图像。 总体一千张,我按9:1分的训练和验证。

getstart里的demo运行train.py也没有labels.jpg的两个文件,权重文件时下载的您上传的度盘文件,只是名字没改,其他都一样。 运行的时候有个这个警告:UserWarning: Detected call of `lr_scheduler.step()` before `optimizer.step()`, 效果如下,感觉也有点问题,我做训练时nc的类型都注意改了的,唯一有改动的是imgsize,我改到了1280. train_batch ![image](https://user-images.githubusercontent.com/91964911/181308551-1462555a-7847-4176-85c6-9885d39ddd4c.png) val_batch_labels ![image](https://user-images.githubusercontent.com/91964911/181308976-d3b5476c-52a7-46c5-92c1-67b7bddc1b8b.png) val_batch_pre ![image](https://user-images.githubusercontent.com/91964911/181309026-003e5686-e68d-4628-9347-3d1a83078ee1.png)

> demo也没有的话,检查下环境依赖的版本和requirements.txt是否保持一致 我解决了训练P,R为0的问题,但还是没有labels的两个文件,大概原因就是16xx系的显卡不支持cuda11.3,换回10.2就可以了 具体可以参考这个 https://blog.csdn.net/qq_52902342/article/details/124261371 ![image](https://user-images.githubusercontent.com/91964911/181462851-18a18644-348c-4d5c-9f73-2c6df91d5715.png)