Wulx2050
Wulx2050
Why not dense unknown boxes? In the "A sample qualitative result" example, apples and oranges are marked as unknown to begin with, but why aren't glasses, human ears, eyebrows, hair,...
在使用 `deepxde` 训练的时候，为了测试一些极端情况，我经常把输入数据设置的极小(接近0)或者超过几百，或者有些写了错误的边界条件、初始条件，这时候训练模型`Train loss`和`Test loss`会出现`nan`，并且立刻停止训练。出现这种情况一般需要截断误差，即如果一旦损失值超过某个界限[lossmax]，就直接截断，取预先设定的值。规范化 PDE或分配损失权重( normalize the PDE, or assign a loss weight) 需要调整很多次，而且如果我有10000个形式不同的PDE问题，难道我要手动规范化10000个 PDE吗？ by Google Translate： When using `deepxde` for training, in order to test some extreme...
Imagen' key discovery is that generic large language models (e.g. T5), pretrained on text-only corpora, are surprisingly effective at encoding text for image synthesis: increasing the size of the language...
How about fastdiff's support for Chinese voice? Did you do experiments?
playable models 1. dalle-mini & craiyon https://github.com/borisdayma/dalle-mini 2. CogView2 https://github.com/THUDM/CogView2 3. 待添加 ----------------------------- No pretrained models 1. imagen https://github.com/lucidrains/imagen-pytorch 2. 文心 ERNIE-ViLG https://wenxin.baidu.com/wenxin/modelbasedetail/ernie_vilg/ 3. 待添加