【Hackathon No.29】add RFC for PixelUnshuffle #44

BrilliantYuKaimin · 2022-03-17T10:12:18Z

增加paddle.nn.PixelUnshuffle设计文档。

shiyutang

方案整体完备，但是实现细节等需要步骤化说明，请修改方案后再提交代码。修改示例参照：https://github.com/PaddlePaddle/community/blob/master/rfcs/APIs/20200301_api_design_for_quantile.md

shiyutang · 2022-03-17T12:37:10Z

rfcs/APIs/20220317_api_design_for_PxielUnshuffle.md

+## 飞桨
+
+飞桨目前没有直接提供`ChannelShuffle`的API，但是也可以通过组合API的方式实现该操作：
+
+```python
+class PixelUnshuffle(paddle.nn.Layer):
+    def __init__(self, downscale_factor):
+        super(PixelUnshuffle, self).__init__()
+        self.factor = downscale_factor
+        self.data_format = data_format
+
+    def forward(self, x):
+        n, c, h, w = x.shape
+        x = paddle.reshape(x, [n, c, h / self.factor, self.factor, w / self.factor, self.factor])
+        x = paddle.transpose(x, [0, 1, 3, 5, 2, 4])
+        x = paddle.reshape(x, [n, c * self.factor * self.factor, h / self.factor, w / self.factor])
+        return x
+```


这部分可以放在飞桨现状

这部分可以放在飞桨现状

完成

shiyutang · 2022-03-17T12:38:03Z

rfcs/APIs/20220317_api_design_for_PxielUnshuffle.md

+- 与PyTorch的结果的一致性；
+- 反向传播的正确性；
+- 错误检查：`downscale_factor`不是正整数或不同时整除宽度和高度时能正确抛出异常。
+


这部分需要修改，请参考这个PR中描述：#40

这部分需要修改，请参考这个PR中描述：#40

完成

shiyutang · 2022-03-17T12:39:22Z

rfcs/APIs/20220317_api_design_for_PxielUnshuffle.md

+## API实现方案
+
+参考`paddle.nn.PixelShuffle`来实现`paddle.nn.PixelUnshuffle`，顺便实现`paddle.nn.functional.pixel_unshuffle`。
+


这部分需要修改，请参考这个PR中描述：#40

这部分需要修改，请参考这个PR中描述：#40

完成

shiyutang · 2022-03-17T12:39:33Z

rfcs/APIs/20220317_api_design_for_PxielUnshuffle.md

+# 四、对比分析
+
+无论是C++实现还是组合API实现，其逻辑都是十分简单的，故考虑使用C++编写新的算子以期取得更高的效率。


这部分需要修改，请参考这个PR中描述：#40

这部分需要修改，请参考这个PR中描述：#40

完成

shiyutang · 2022-03-17T12:39:57Z

rfcs/APIs/20220317_api_design_for_PxielUnshuffle.md

+其中的`F.pixel_unshuffle`是由C++实现的，主要代码为：
+
+```c++
+Tensor pixel_unshuffle(const Tensor& self, int64_t downscale_factor) {
+  TORCH_CHECK(self.dim() >= 3,
+              "pixel_unshuffle expects input to have at least 3 dimensions, but got input with ",
+              self.dim(), " dimension(s)");
+  TORCH_CHECK(
+      downscale_factor > 0,
+      "pixel_unshuffle expects a positive downscale_factor, but got ",
+      downscale_factor);
+  int64_t c = self.size(-3);
+  int64_t h = self.size(-2);
+  int64_t w = self.size(-1);
+  constexpr auto NUM_NON_BATCH_DIMS = 3;
+  const auto self_sizes_batch_end = self.sizes().end() - NUM_NON_BATCH_DIMS;
+
+  TORCH_CHECK(h % downscale_factor == 0,
+             "pixel_unshuffle expects height to be divisible by downscale_factor, but input.size(-2)=", h,
+             " is not divisible by ", downscale_factor)
+  TORCH_CHECK(w % downscale_factor == 0,
+             "pixel_unshuffle expects width to be divisible by downscale_factor, but input.size(-1)=", w,
+             " is not divisible by ", downscale_factor)
+  int64_t downscale_factor_squared = downscale_factor * downscale_factor;
+  int64_t oc = c * downscale_factor_squared;
+  int64_t oh = h / downscale_factor;
+  int64_t ow = w / downscale_factor;
+
+  std::vector<int64_t> added_dims_shape(
+      self.sizes().begin(), self_sizes_batch_end);
+  added_dims_shape.insert(
+      added_dims_shape.end(), {c, oh, downscale_factor, ow, downscale_factor});
+  const auto input_reshaped = self.reshape(added_dims_shape);
+
+  std::vector<int64_t> permutation(self.sizes().begin(), self_sizes_batch_end);
+  std::iota(permutation.begin(), permutation.end(), 0);
+  permutation.insert(permutation.end(), {-5 /* c */, -3 /* 1st downscale_factor */, -1 /*2nd downscale_factor */,
+                                         -4 /* oh */, -2 /* ow */});
+  const auto input_permuted = input_reshaped.permute(permutation);
+
+  std::vector<int64_t> final_shape(self.sizes().begin(), self_sizes_batch_end);
+  final_shape.insert(final_shape.end(), {oc, oh, ow});
+  return input_permuted.reshape(final_shape);
+}
+
+}} // namespace at::native
+
+```
+
+## TensorFlow
+
+TensorFlow目前没有直接提供`ChannelShuffle`的API，但是也可以通过组合API的方式实现该操作：
+
+```python
+def shuffle_unit(self, x, downscale_factor):
+    with tf.variable_scope('pixel_unshuffle_unit'):
+        n, h, w, c = x.get_shape().as_list()
+        x = tf.reshape(x, shape=tf.convert_to_tensor([tf.shape(x)[0], h / downscale_factor, downscale_factor, w / downscale_factor, downscale_factor, c]))
+        x = tf.transpose(x, tf.convert_to_tensor([0, 1, 3, 5, 2, 4]))
+        x = tf.reshape(x, shape=tf.convert_to_tensor([tf.shape(x)[0], h / downscale_factor, w / downscale_factor, c * downscale_factor * downscale_factor]))
+
+```


这部分需要增加内容，请参考这个PR中描述：#40

这部分需要增加内容，请参考这个PR中描述：#40

完成

paddle-bot-old · 2022-03-18T08:33:28Z

PR格式检查通过，你的PR将接受Paddle专家以及开源社区的review，请及时关注PR动态。
The format inspection passed. Your PR will be reviewed by experts of Paddle and developers from the open-source community. Stay tuned.

paddle-bot-old · 2022-03-18T08:33:33Z

你的PR有最新反馈，请及时修改。
There’s the latest feedback about your PR. Please check.

shiyutang

参考#40 进行同步修改

shiyutang · 2022-03-18T09:01:52Z

rfcs/APIs/20220317_api_design_for_PxielUnshuffle.md

+
+}} // namespace at::native
+


需要进一步增加对torch方案的步骤化描述

shiyutang · 2022-03-18T09:04:22Z

rfcs/APIs/20220317_api_design_for_PxielUnshuffle.md

+- 是否同时支持CPU和GPU平台；
+- 测试不同张量类型下的表现；
+- 对全部入参进行参数有效性和边界值测试，确定每个入参都可以正确生效；
+- 前向计算的正确性（与组合API实现比较、与PyTorch比较）；


本地torch，CI中使用numpy实现的算法进行对比

shiyutang · 2022-03-18T09:05:16Z

rfcs/APIs/20220317_api_design_for_PxielUnshuffle.md

+- 当传入的`downscale_factor`不合法（不是正整数、不同时整除高度和宽度）时会抛出异常并有友好的提示。
+


错误检查：输入维度，downscale_factor > 0等

按照评审意见修改了代码

shiyutang

LGTM

This reverts commit 4a82d48.

shiyutang · 2022-03-22T12:03:56Z

实现内容可以存放在paddle/phi/kernels/cpu(gpu)/ 下，注册则在fluid/oprerator下。如下图所示：

c++ op开发方式可以参见开发指南：https://www.paddlepaddle.org.cn/documentation/docs/zh/develop/dev_guides/api_contributing_guides/new_cpp_op_cn.html

shiyutang · 2022-03-22T12:04:24Z

请对文档中的实现内容位置进行删除或者对应修改

BrilliantYuKaimin · 2022-03-22T12:15:17Z

请对文档中的实现内容位置进行删除或者对应修改

因为cpu和gpu的kernel是一样的，所以参考pixel_shuffle把实现放在paddle/phi/kernels/impl/pixel_unshuffle_kernel_impl.h中。

shiyutang · 2022-03-23T09:39:49Z

如果实现和设备无关，应当放在kernel根目录下，如下所示：

shiyutang · 2022-03-23T09:40:18Z

请提交一个新的PR对相应内容进行修改

paddle-bot-old · 2022-03-24T02:48:22Z

很抱歉，经过我们的反复讨论，你的PR暂未达到合入标准，请阅读飞桨社区贡献指南，你可以重新提交新的PR，我们先将此PR关闭，感谢你的贡献。
Sorry to inform you that through our discussion, your PR fails to meet the merging standard (Reference: Paddle API Design Standard Doc). You can also submit an new one. Thank you.

Create 20220317_api_design_for_PxielUnshuffle.md

e1f229a

paddle-bot-old bot added contributor status: proposed labels Mar 17, 2022

Update 20220317_api_design_for_PxielUnshuffle.md

4e92f4b

BrilliantYuKaimin mentioned this pull request Mar 17, 2022

【PaddlePaddle Hackathon 第二期】任务总览 PaddlePaddle/Paddle#40234

Closed

shiyutang requested changes Mar 17, 2022

View reviewed changes

Update 20220317_api_design_for_PxielUnshuffle.md

c4ef309

yang131313 added status: open review and removed status: proposed labels Mar 18, 2022

yang131313 added status: revision and removed status: open review labels Mar 18, 2022

shiyutang requested changes Mar 18, 2022

View reviewed changes

Update 20220317_api_design_for_PxielUnshuffle.md

d47c5bc

按照评审意见修改了代码

yang131313 assigned DDDivano and shiyutang Mar 21, 2022

dingjiaweiww mentioned this pull request Mar 21, 2022

【Hackathon No.29】add RFC for PixelUnshuffle #43

Closed

shiyutang approved these changes Mar 22, 2022

View reviewed changes

shiyutang merged commit 4a82d48 into PaddlePaddle:master Mar 22, 2022

shiyutang added a commit that referenced this pull request Mar 22, 2022

Revert "【Hackathon No.29】add RFC for PixelUnshuffle (#44)"

7234c75

This reverts commit 4a82d48.

shiyutang mentioned this pull request Mar 22, 2022

Revert "【Hackathon No.29】add RFC for PixelUnshuffle" #59

Closed

dingjiaweiww added status: not progressed and removed status: revision labels Mar 24, 2022

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

【Hackathon No.29】add RFC for PixelUnshuffle #44

【Hackathon No.29】add RFC for PixelUnshuffle #44

BrilliantYuKaimin commented Mar 17, 2022

shiyutang left a comment

shiyutang Mar 17, 2022

BrilliantYuKaimin Mar 17, 2022

shiyutang Mar 17, 2022

BrilliantYuKaimin Mar 17, 2022

shiyutang Mar 17, 2022

BrilliantYuKaimin Mar 17, 2022

shiyutang Mar 17, 2022

BrilliantYuKaimin Mar 17, 2022

shiyutang Mar 17, 2022

BrilliantYuKaimin Mar 17, 2022

paddle-bot-old bot commented Mar 18, 2022

paddle-bot-old bot commented Mar 18, 2022

shiyutang left a comment

shiyutang Mar 18, 2022

BrilliantYuKaimin Mar 18, 2022

shiyutang Mar 18, 2022

BrilliantYuKaimin Mar 18, 2022

shiyutang Mar 18, 2022

BrilliantYuKaimin Mar 18, 2022

shiyutang left a comment

shiyutang commented Mar 22, 2022

shiyutang commented Mar 22, 2022

BrilliantYuKaimin commented Mar 22, 2022

shiyutang commented Mar 23, 2022

shiyutang commented Mar 23, 2022

paddle-bot-old bot commented Mar 24, 2022

		## API实现方案

		参考`paddle.nn.PixelShuffle`来实现`paddle.nn.PixelUnshuffle`，顺便实现`paddle.nn.functional.pixel_unshuffle`。

		# 四、对比分析

		无论是C++实现还是组合API实现，其逻辑都是十分简单的，故考虑使用C++编写新的算子以期取得更高的效率。

		- 当传入的`downscale_factor`不合法（不是正整数、不同时整除高度和宽度）时会抛出异常并有友好的提示。

【Hackathon No.29】add RFC for PixelUnshuffle #44

【Hackathon No.29】add RFC for PixelUnshuffle #44

Conversation

BrilliantYuKaimin commented Mar 17, 2022

shiyutang left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

paddle-bot-old bot commented Mar 18, 2022

paddle-bot-old bot commented Mar 18, 2022

shiyutang left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

shiyutang left a comment

Choose a reason for hiding this comment

shiyutang commented Mar 22, 2022

shiyutang commented Mar 22, 2022

BrilliantYuKaimin commented Mar 22, 2022

shiyutang commented Mar 23, 2022

shiyutang commented Mar 23, 2022

paddle-bot-old bot commented Mar 24, 2022