MMXuan

Results 27 issues of MMXuan

在pytorch1.10.0用script模式导出的模型中有if和loop结构,模型会根据不同的输入参数值会有不同的运行路径,能成功导出onnx模型并用onnxruntime加载,但是在转mnn时候报如下错误 ![image](https://user-images.githubusercontent.com/6283983/157147431-c4c0180e-5f9d-4ce2-bfaa-70b072928438.png)

**1. 使用环境(environment)** - OS: Mac/Ubuntu/Centos/Docker - OS Version: **2. Github版本** - branch: - commit(optional): **3. 详细描述bug 情况 (Describe the bug)** A clear and concise description of what the bug is....

**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** 使用下面命令转ppmatting的模型为 onnx时报错 paddle2onnx --model_dir . --model_filename model.pdmodel --params_filename model.pdiparams --save_file model.onnx --enable_dev_version True --opset_version 15 [ERROR][Paddle2ONNX][pool2d: pool2d_1.tmp_0] Adaptive only support static input shape. [Paddle2ONNX] Due to the operator:...

onnx模型里面有LOOP和IF算子,支持这种结构吗?

why add this sum make all class weight the same ![image](https://user-images.githubusercontent.com/6283983/105962017-38ea6100-60ba-11eb-9b60-9dc4c07a1fe8.png)

this repo is a very good accuracy rate benchmark reference for all model on imagenet classification task, but when do real project deployment the model's inference time cost is also...

enhancement

看train.py里直接就是rpn和roi两阶段一起直接训练的,这里没有采用论文里面分两个阶段训练的模式是一次训练能直接达到效果吗?

QFL的描述很相似的感觉,在heatmap的中心点周围去增加惩罚 ![image](https://user-images.githubusercontent.com/6283983/100306275-fb942500-2fdd-11eb-8bdc-7927d4b7f3f4.png) ![image](https://user-images.githubusercontent.com/6283983/100306298-08187d80-2fde-11eb-9274-89f959b4d5fa.png)

![image](https://user-images.githubusercontent.com/6283983/42497251-4af39e7c-845b-11e8-8e17-002ea32765c5.png)