xxddccaa issues

Results 22 issues of


                                            xxddccaa

PaddleClas MultiLabels模型导出的onnx模型推理有BUG

MultiLabels模型导出的onnx模型推理与 paddle模型相差非常大。 1 代码：PaddleClas develop分支 2 环境：docker images paddlepaddle/paddle:2.1.0-gpu-cuda11.2-cudnn8 下面是我操作的过程： # 导出到paddle推理模型 python3 tools/export_model.py -c ./ppcls/configs/quick_start/professional/MobileNetV1_multilabel_intent_eval.yaml -o Arch.pretrained="./output/k1000_focal_2/MobileNetV2/latest" ![image](https://user-images.githubusercontent.com/25652821/181870448-eff7df78-d052-47ba-b19b-c2b03e24eec0.png) # 安装必要的包 python -m pip install -i https://pypi.tuna.tsinghua.edu.cn/simple --upgrade pip...

paddleclas中multilabels模型转 onnx后精度损失太多

看了一些issues都不能解决我的问题： https://github.com/PaddlePaddle/Paddle2ONNX/issues/451 https://github.com/PaddlePaddle/Paddle2ONNX/issues/772 这是我的issues： https://github.com/PaddlePaddle/PaddleClas/issues/2174 对于同一张图输入，paddle模型输出后sigmod是（最大的两个数值是0.10090836, 0.96171629）： ![image](https://user-images.githubusercontent.com/25652821/181877214-62c72705-a1e5-4c0a-978d-f9ea289233de.png) 而onnx模型里面带了sigmod，输出就是想要的数值（最大的两个数值是0.00439452, 0.98361409）： ![image](https://user-images.githubusercontent.com/25652821/181877225-22ace330-f9f4-462b-b09b-617f8e0ff555.png) 想不出来是哪里出了问题，是paddle转onnx之后精度差这么多吗？

[Feature] how to use gradio_inpainting.py?

### What's the feature? I fail to use `python demo/gradio_inpainting.py` , I couldn't find a tutorial on how to use this. thx~ $ python demo/gradio_inpainting.py ==model_name aot_gan ==meta_file_dir /ssd/xiedong/workplace/mmagic Editable...

kind/enhancement

It is now 2023, and most people dare not include code for video object detection. 如今已经2023年，关于视频目标检测的大都不敢放代码。

Why on earth is it

where is "cfg file of yolov7-w6-face"

thank you for your project. i don't find yolov7-w6-face.yaml in your cfg directory, can you update, thx~

还可以的

Epoch [1/30], Loss: 0.0000 Accuracy on test set: 81.23% 0.009939057285945933 Epoch [2/30], Loss: 0.0245 Accuracy on test set: 76.59% 0.009757729755661013 Epoch [3/30], Loss: 0.0623 Accuracy on test set: 77.08% 0.009460482294732423...

Hope to add upscaler: UpscalerData to API

i use api，this greatly increases flexibility: ``` def roop_api(_: gr.Blocks, app: FastAPI): @app.post("/roop/image") async def roop_image( source_image: str = Body("",title="source face image"), target_image: str = Body("",title="target image"), face_index: list[int] =...

xxddccaa