From 75997fcd601cbf09dbf0e546f084d91f508bffe2 Mon Sep 17 00:00:00 2001
From: Range King <RangeKingHZ@gmail.com>
Date: Sun, 23 Oct 2022 10:53:49 +0800
Subject: [PATCH] [Docs] Refine config.md (#191)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
* Refine __delete__ example in config.md
* Add missing docstring in cspnext.py
* Refine docstring of cspnext
* Correct the description of param_scheduler
* Refine description param_scheduler in config.md
* Update docs/zh_cn/user_guides/config.md
Co-authored-by: Haian Huang(深度眸) <1286304229@qq.com>
Co-authored-by: Haian Huang(深度眸) <1286304229@qq.com>
---
docs/en/user_guides/config.md | 15 +++++++++------
docs/zh_cn/user_guides/config.md | 15 +++++++++------
mmyolo/models/backbones/cspnext.py | 5 +++--
mmyolo/models/necks/cspnext_pafpn.py | 2 +-
4 files changed, 22 insertions(+), 15 deletions(-)
diff --git a/docs/en/user_guides/config.md b/docs/en/user_guides/config.md
index 0800d144..56edadf5 100644
--- a/docs/en/user_guides/config.md
+++ b/docs/en/user_guides/config.md
@@ -286,7 +286,7 @@ optim_wrapper = dict( # Optimizer wrapper config
constructor='YOLOv5OptimizerConstructor') # The constructor for YOLOv5 optimizer
```
-`param_scheduler` is the field that configures methods of adjusting optimization hyperparameters such as learning rate and momentum. Users can combine multiple schedulers to create a desired parameter adjustment strategy. Find more in the [parameter scheduler tutorial](https://mmengine.readthedocs.io/en/latest/tutorials/param_scheduler.html). In MMYOLO, no parameter optimizer is introduced.
+`param_scheduler` is the field that configures methods of adjusting optimization hyperparameters such as learning rate and momentum. Users can combine multiple schedulers to create a desired parameter adjustment strategy. Find more in the [parameter scheduler tutorial](https://mmengine.readthedocs.io/en/latest/tutorials/param_scheduler.html). In YOLOv5, parameter scheduling is complex to implement and difficult to implement with `param_scheduler`. So we use `YOLOv5ParamSchedulerHook` to implement it (see next section), which is simpler but less versatile.
```python
param_scheduler = None
@@ -387,26 +387,29 @@ If you wish to inspect the config file, you may run `mim run mmdet print_config
Sometimes, you may set `_delete_=True` to ignore some of the fields in base configs.
You may refer to the [mmengine config tutorial](https://mmengine.readthedocs.io/en/latest/tutorials/config.html) for a simple illustration.
-In MMYOLO, for example, to change the backbone of YOLOv5 with the following config.
+In MMYOLO, for example, to change the backbone of RTMDet with the following config.
```python
model = dict(
type='YOLODetector',
data_preprocessor=dict(...),
backbone=dict(
- type='YOLOv5CSPDarknet',
+ type='CSPNeXt',
+ arch='P5',
+ expand_ratio=0.5,
deepen_factor=deepen_factor,
widen_factor=widen_factor,
- norm_cfg=dict(type='BN', momentum=0.03, eps=0.001),
+ channel_attention=True,
+ norm_cfg=dict(type='BN'),
act_cfg=dict(type='SiLU', inplace=True)),
neck=dict(...),
bbox_head=dict(...))
```
-The `_delete_=True` would replace all old keys in the `backbone` field with new keys. For example, `YOLOv5` uses `YOLOv5CSPDarknet`, it is necessary to replace the backbone with `YOLOv6EfficientRep`. Since `YOLOv5CSPDarknet` and `YOLOv6EfficientRep` have different fields, you need to use `_delete_=True` to replace all old keys in the `backbone` field.
+If you want to change `CSPNeXt` to `YOLOv6EfficientRep` for the RTMDet backbone, because there are different fields (`channel_attention` and `expand_ratio`) in `CSPNeXt` and `YOLOv6EfficientRep`, you need to use `_delete_=True` to replace all the old keys in the `backbone` field with the new keys.
```python
-_base_ = '../yolov5/yolov5_s-v61_syncbn_8xb16-300e_coco.py'
+_base_ = '../rtmdet/rtmdet_l_syncbn_8xb32-300e_coco.py'
model = dict(
backbone=dict(
_delete_=True,
diff --git a/docs/zh_cn/user_guides/config.md b/docs/zh_cn/user_guides/config.md
index 17b64a6c..355ef5d1 100644
--- a/docs/zh_cn/user_guides/config.md
+++ b/docs/zh_cn/user_guides/config.md
@@ -287,7 +287,7 @@ optim_wrapper = dict( # 优化器封装的配置
```
-`param_scheduler` 字段用于配置参数调度器（Parameter Scheduler）来调整优化器的超参数（例如学习率和动量）。 用户可以组合多个调度器来创建所需的参数调整策略。 在[参数调度器教程](https://mmengine.readthedocs.io/zh_CN/latest/tutorials/param_scheduler.html) 和参数调度器 API 文档 中查找更多信息。在 MMYOLO 中，未引入任何参数调度器。
+`param_scheduler` 字段用于配置参数调度器（Parameter Scheduler）来调整优化器的超参数（例如学习率和动量）。 用户可以组合多个调度器来创建所需的参数调整策略。 在[参数调度器教程](https://mmengine.readthedocs.io/zh_CN/latest/tutorials/param_scheduler.html) 和参数调度器 API 文档 中查找更多信息。在 YOLOv5 中，参数调度实现比较复杂，难以通过 `param_scheduler` 实现。所以我们采用了 `YOLOv5ParamSchedulerHook` 来实现（见下节），这样做更简单但是通用性较差。
```python
param_scheduler = None
@@ -386,26 +386,29 @@ _base_ = [
有时，您也许会设置 `_delete_=True` 去忽略基础配置文件里的一些域内容。 您也许可以参照 [MMEngine 配置文件教程](https://mmengine.readthedocs.io/en/latest/tutorials/config.html) 来获得一些简单的指导。
-在 MMYOLO 里，例如为了改变 YOLOv5 的主干网络的某些内容：
+在 MMYOLO 里，例如为了改变 RTMDet 的主干网络的某些内容：
```python
model = dict(
type='YOLODetector',
data_preprocessor=dict(...),
backbone=dict(
- type='YOLOv5CSPDarknet',
+ type='CSPNeXt',
+ arch='P5',
+ expand_ratio=0.5,
deepen_factor=deepen_factor,
widen_factor=widen_factor,
- norm_cfg=dict(type='BN', momentum=0.03, eps=0.001),
+ channel_attention=True,
+ norm_cfg=dict(type='BN'),
act_cfg=dict(type='SiLU', inplace=True)),
neck=dict(...),
bbox_head=dict(...))
```
-基础配置的 `YOLOv5` 使用 `YOLOv5CSPDarknet`，在需要将主干网络改成 `YOLOv6EfficientRep` 的时候，因为 `YOLOv5CSPDarknet` 和 `YOLOv6EfficientRep` 中有不同的字段，需要使用 `_delete_=True` 将新的键去替换 `backbone` 域内所有老的键。
+如果想把 RTMDet 主干网络的 `CSPNeXt` 改成 `YOLOv6EfficientRep`，因为 `CSPNeXt` 和 `YOLOv6EfficientRep` 中有不同的字段(`channel_attention` 和 `expand_ratio`)，这时候就需要使用 `_delete_=True` 将新的键去替换 `backbone` 域内所有老的键。
```python
-_base_ = '../yolov5/yolov5_s-v61_syncbn_8xb16-300e_coco.py'
+_base_ = '../rtmdet/rtmdet_l_syncbn_8xb32-300e_coco.py'
model = dict(
backbone=dict(
_delete_=True,
diff --git a/mmyolo/models/backbones/cspnext.py b/mmyolo/models/backbones/cspnext.py
index 2ef0c896..f12e37a4 100644
--- a/mmyolo/models/backbones/cspnext.py
+++ b/mmyolo/models/backbones/cspnext.py
@@ -28,12 +28,13 @@ class CSPNeXt(BaseBackbone):
frozen_stages (int): Stages to be frozen (stop grad and set eval
mode). -1 means not freezing any parameters. Defaults to -1.
plugins (list[dict]): List of plugins for stages, each dict contains:
-
- - cfg (dict, required): Cfg dict to build plugin.
+ - cfg (dict, required): Cfg dict to build plugin.Defaults to
- stages (tuple[bool], optional): Stages to apply plugin, length
should be same as 'num_stages'.
use_depthwise (bool): Whether to use depthwise separable convolution.
Defaults to False.
+ expand_ratio (float): Ratio to adjust the number of channels of the
+ hidden layer. Defaults to 0.5.
arch_ovewrite (list): Overwrite default arch settings.
Defaults to None.
channel_attention (bool): Whether to add channel attention in each
diff --git a/mmyolo/models/necks/cspnext_pafpn.py b/mmyolo/models/necks/cspnext_pafpn.py
index 710d85dd..a4dc846e 100644
--- a/mmyolo/models/necks/cspnext_pafpn.py
+++ b/mmyolo/models/necks/cspnext_pafpn.py
@@ -27,7 +27,7 @@ class CSPNeXtPAFPN(BaseYOLONeck):
use_depthwise (bool): Whether to use depthwise separable convolution in
blocks. Defaults to False.
expand_ratio (float): Ratio to adjust the number of channels of the
- hidden layer. Default: 0.5
+ hidden layer. Defaults to 0.5.
upsample_cfg (dict): Config dict for interpolate layer.
Default: `dict(scale_factor=2, mode='nearest')`
conv_cfg (dict, optional): Config dict for convolution layer.