-
我想将ppyolov2转换为onnx模型,过程中遇到如下问题:
我首先执行如下命令将ppyolov2成功导出模型
python3 tools/export_model.py -c configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco.yml -o weights=./models/ppyolov2_r50vd_dcn_365e_coco.pdparams
…
-
系统环境为:win10+cuda11.2+cudnn8.1+TensorRT8.0.1.6+Python3.8
当使用pdx.deploy.Predictor初始化加载模型时,use_trt=False时能正常加载,=True时会报错。代码及报错如下图
![image](https://user-images.githubusercontent.com/31558283/145660776-b…
-
win10 python3.6.4 paddleDetection2.1
无法运行ppyolov2_r50vd_dcn_365e_coco模型,报错如下,但是配置文件本来就没有那个参数,不知道怎么解决
![image](https://user-images.githubusercontent.com/86935392/126073947-a74a625d-5aaa-4c8c-993b-95…
itues updated
3 years ago
-
环境:Win10+Python3.8+Cuda11.2+Cudnn8.2+TensorRT8.0.1.6
使用的是在官网PaddleInference板块里下载的预测库
![image](https://user-images.githubusercontent.com/31558283/171304104-25080cda-fd70-460e-8ff8-79a643de7cb9.png)
…
-
在i9-10000k 内存128g,显卡3080 。windows10平台,cuda11。发现同样1920*1080图片,ppyolov2,训练模型608 单张检测耗时需要接近300ms,而采用yolov5训练模型1280,最慢是50ms左右。
-
进行单机双卡训练时,不定时间异常中止,进行排查发现训练过程中内存泄漏。
使用代码:
```py
import paddle
import paddle.nn as nn
import paddle.optimizer as opt
import paddle.distributed as dist
import paddlex as pdx
from paddlex impor…
-
想问一下检测问题中使用ppyolov2时这里输入是不同的尺寸均匀抽取,那如果想自己计算anchor适应自己的数据集的时候,应该按照哪个尺寸去计算,对于这种均与抽取的操作是从ppyolo才开始出现的吗,我记得v3v4里面好像都没有用过。还是说用这种均匀抽取时就只能使用默认的coco的anchor
-
在aistudio上训练ppyolov2,开始时可以正常训练,后面就变成nan了。
![MCY Q05{UG@{6((@8HBV~_1](https://user-images.githubusercontent.com/71364549/117782238-04b73c80-b274-11eb-8990-1f1c2de9507a.png)
我按照文档将学习率除以8,然后将batch siz…
-
根据官方教程跑了ppyolov2模型,利用roadsign数据集训练,Loss训练到最后一直卡在10几不下降,AP不到0.1
单卡训练,学习率也缩小了10倍
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'./_base_/ppyolov2_r50vd_dcn.yml',
'./_base…
-
问题类型:模型训练
**PaddleX版本**
您使用的PaddleX版本
**问题描述**
(paddlex) C:\Users\zzy\Desktop\paddle\tutorials\train\object_detection>python ppyolov2.py
E:\anaconda3\envs\paddlex\lib\site-packages\pa…