sunjian2015

memery is full used at the beginning of validation phase

7

When I trained segformer model, there was no problem in the training phase, but during validation, the memory was fully occupied when reading pictures, then the process was killed. Has...

可视化时报错

2

``` [ERROR 2022-04-20 11:11:37,329 lib.py:537] Unexpected error: 'LogReader' object has no attribute 'data_manager' [ERROR 2022-04-20 11:11:39,331 lib.py:537] Unexpected error: 'LogReader' object has no attribute 'data_manager' Traceback (most recent call last):...

Does anyone knows what's the difference between clean version and original version? which version performance is better?

请问忽略小脸该如何处理呢？

1

如题，需要在代码的哪些地方做一些改动？

类似的还有这个问题 WARNING: Ignoring corrupted image and/or label /path/to/widerface/train/12_Group_Team_Organized_Grou p_12_Group_Team_Organized_Group_12_994.jpg: non-normalized or out of bounds coordinate labels 请问有遇到这种问题的吗？该如何解决，或者可以直接忽略吗？

感谢您参与 X2Paddle 社区！ 问题模版为了 X2Paddle 能更好的迭代，例如新功能发布、 RoadMaps 和错误跟踪. :smile_cat: ## 问题描述 pytorch nn.linear bias 设置为 False 时，被转换为 None，转换后代码 paddle.add 提示不能为 None，如下图 ## 具体信息 - 转换模型后用处 - [x] 使用 Paddle 框架/...

I have tested HRNet-W16 and Lite-HRNet model on my segmentation task, but the conclusion I obtained is contrary to the conclusion in the paper.

如题，看了下代码，没有找到图中的 Channel wise attention 呀，而且，ParC-block 也只有一次 ParC-H 和 ParC-V 的 concatenate。求指教~

### 📚 The doc issue 2D 人脸关键点检测，和 105，106 重叠的是哪两个？ 看不出来啊 ### Suggest a potential alternative/fix 建议都写上

推理时中文唇形同步不准确

3

如题，也试过用中文数据训练，但是也是同步不好，请问有什么办法可以解决吗？而且嘴唇抖动明显