Windaway

Add octconv layers and obj coord preprocessing ops.

1

Add octconv layers and obj coord preprocessing ops. ### Checklist - [x] I've tested that my changes are compatible with the latest version of Tensorflow. - [x] I've read the...

The network class in the old code does not set the default value, so there will be problem with the training. Will the rework NAN come recently? I have implement...

The implement uses init_weight to load the pretrained model. However, you use nn.Sequential to repackage the nn.Modules. So, the code has two sets of parmaters, and you only load weight...

Will give more detail about the network?

4

I have reimplement the paper. How ever my implement cost a lot of memory with a just resnet50 backbone, which is very strange. The paper said one P100 can hold...

### New Issue Checklist - [×] I have read the [Contribution Guidelines](https://github.com/tensorlayer/tensorlayer/blob/master/CONTRIBUTING.md) - [×] I searched for [existing GitHub issues](https://github.com/tensorlayer/tensorlayer/issues) ### Issue Description Pytorch后端模型定义NHWC和NCHW数据格式主要是定以数据和模型后传到设备时用.to("cuda:0", memory_format=torch.channels_last)确定。 TLX目前做法是pytorch依据nhwc格式时，全部转NCHW然后处理完转回来，这潜在是让模型用NCHW格式计算。对纯GPU应用时问题不大，但是对于一些NHWC友好的设备部署，比如未来的Mindspore，由于多次nhwc nchw切换，性能有损失。 这里可能需要框架对于pytorch这里nhwc支持改成全局变量，即输入时数据做nchw-nhwc，模型转nhwc然后计算即可。 不过Pytorch本身GPU NHWC支持稀烂，倒不是很急。

手机为Realme GTNEO，安卓13系统。软件版本0.8.1。测试的歌单如下，约928KB。 [listen1_backup.zip](https://github.com/listen1/listen1_mobile/files/11264383/listen1_backup.zip)

添加Anaconda源

2

深度学习很依赖这些源 外加第三方的forge和pytorch源

Invalid issue

Since Distinctions-646 does not provide trimaps for the test set, ViTMatte states that it randomly generate trimap according to alpha mattes using erosion operations with random kernel size in [1,...