CashBai

Results 17 issues of CashBai

问题如上,遇到项目对预测速度有要求,需要用到tensorrt加速部署,x86平台,rtx3080ti

### 问题确认 Search before asking - [X] 我已经搜索过问题,但是没有找到解答。I have searched the question and found no related answer. ### 请提出你的问题 Please ask your question 如题,没有找到数据增强算子对应的输入参数说明及文档

status/close

![image](https://github.com/user-attachments/assets/c479c3ad-b53c-4d22-aa41-fd1f9ab7481e) ![image](https://github.com/user-attachments/assets/d96dcd61-2937-42ca-96a4-acf3a15aa0f2) 安装模式用的FD文档中的Conda指令“conda config --add channels conda-forge && conda install cudatoolkit=11.2 cudnn=8.2” PPYOLOE+的模型是用的Paddledetection训练的,develop版本 ![image](https://github.com/user-attachments/assets/05a93a1c-a69c-4cae-b8ab-34876257981e)

![image](https://github.com/user-attachments/assets/8484c600-f4a4-4485-b197-ec1275c18549) Deeplabv3p的模型

我有一个数电池数量的数据集,用到了PP-YOLOE_plus-L模型 在训练的后半程,所有的评估bbox ap都是0.392 我一张图片中有256个目标,256*0.392=100 所以我想问一下,是不是训练过程中的验证集评估中是有目标数量上限的,且上限是100 ![Image](https://github.com/user-attachments/assets/4be3b621-59c6-4789-8792-76190e493416) ![Image](https://github.com/user-attachments/assets/7adcfe79-1204-4730-bab0-bfbe0bae513e)

使用PP-YOLOE_plus_SOD-L,运行到验证集评估的阶段就开始报错 ``` [05/21 14:21:36] ppdet.engine.callbacks INFO: Epoch: [0] [380/396] learning_rate: 0.000025 loss: 0.864219 loss_cls: 0.473560 loss_iou: 0.049636 loss_dfl: 0.528526 loss_l1: 0.049658 eta: 0:32:15 batch_cost: 0.2519 data_cost: 0.0014 ips: 7.9396 images/s,...

平台为win10 训练时的cmd里记录如下: ``` LAUNCH INFO 2024-12-23 13:55:59,670 ----------- Configuration ---------------------- LAUNCH INFO 2024-12-23 13:55:59,671 auto_parallel_config: None LAUNCH INFO 2024-12-23 13:55:59,671 auto_tuner_json: None LAUNCH INFO 2024-12-23 13:55:59,671 devices: 0,1 LAUNCH INFO...