Chenxiao Niu

Hi, Thanks for your great work, but can you please also provide the build file? Thanks again.

I know that the resnet20 implementation has some modifications, but why the parameters count is different from torchvision by a large margin (0.79M vs. 11.6M). It is a huge difference.

Is it possible to add random copy-paste support in the future?

The following changes made: * compute segments from bounding box. * make img_id and anno_id independent with voc annotation, the original way may cause some bugs if voc annotation file...

bug fix of [this issue](https://github.com/Hironsan/bertsearch/issues/33#issue-1199976419) ### Main changes: 1. use latest flask version for docker-compose start problem 2. use clip-as-service as bert embedding backend since the original repo is deprecated...

[MLU] add int64 support for mlu one_hot_v2

1

### PR types Function optimization ### PR changes OPs ### Describe add int64 support for mlu one_hot_v2 kernel

Hi, Thanks for you awesome work, it helps a lot in developing in docker container; however, I have met some problem: I added a docker directory into vscode's workspace correctly,...

grid_sample算子导致paddle2onnx强制使用opset_version 16

3

**请将下面信息填写完整，便于我们快速解决问题，谢谢！** **问题描述** RT，我在尝试使用PaddleDetection的rt-detr模型转换onnx，但是在paddle2onnx时会因为存在grid_sample算子导致工具强制走opset_version 16。 然而我们后面对onnx模型的使用强以来与opset_version 11。有什么办法可以帮忙绕过该算子吗？ 转换方式：[rtdetr](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/rtdetr) **更多信息 :** - 用于部署的推理引擎: 厂家自定的引擎 - 为什么需要转换为ONNX格式：推理引擎依赖onnx parser - Paddle2ONNX版本: 1.0.5 - 你的联系方式(Email/Wechat/Phone): **报错截图** [参考log](https://github.com/PaddlePaddle/PaddleDetection/issues/8114#issuecomment-1552818223) **其他信息**