Spaces:
Runtime error
Runtime error
Hanxiao Xiang
commited on
Commit
·
b328990
1
Parent(s):
0521061
upload
Browse filesThis view is limited to 50 files because it contains too many changes.
See raw diff
- .idea/.gitignore +8 -0
- .idea/Forecast4Muses.iml +14 -0
- .idea/deployment.xml +21 -0
- .idea/inspectionProfiles/Project_Default.xml +75 -0
- .idea/inspectionProfiles/profiles_settings.xml +6 -0
- .idea/misc.xml +4 -0
- .idea/modules.xml +8 -0
- .idea/vcs.xml +6 -0
- Model/Model6/Model6_0_ClothesDetection/__init__.py +0 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/__init__.py +0 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/_base_/__init__.py +0 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/_base_/default_runtime.py +34 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/_base_/det_p5_tta.py +57 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/custom_dataset/.ipynb_checkpoints/__init__.py +0 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/custom_dataset/.ipynb_checkpoints/yolov5_s-v61_syncbn_fast_1xb32-100e_cat-checkpoint.py +135 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/custom_dataset/.ipynb_checkpoints/yolov6_s_fast-checkpoint.py +124 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/custom_dataset/[A]dataset_split.sh +5 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/custom_dataset/__init__.py +0 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/custom_dataset/yolov5_s-v61_syncbn_fast_1xb32-100e_cat.py +135 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/custom_dataset/yolov6_s_fast.py +124 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/custom_dataset/yolov7_l_syncbn_fast_6x16b-100e_coco.py +489 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/custom_dataset/yolov7_l_syncbn_fast_6x32b-100e_coco.py +554 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/custom_dataset/yolov7_l_syncbn_fast_8x16b-300e_coco.py +472 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/README.md +118 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/__init__.py +0 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/crowdhuman/__init__.py +0 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/crowdhuman/yolov5_s-v61_8xb16-300e_ignore_crowdhuman.py +63 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/crowdhuman/yolov5_s-v61_fast_8xb16-300e_crowdhuman.py +47 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/metafile.yml +178 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/voc/__init__.py +0 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/voc/yolov5_l-v61_fast_1xb32-50e_voc.py +25 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/voc/yolov5_m-v61_fast_1xb64-50e_voc.py +17 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/voc/yolov5_n-v61_fast_1xb64-50e_voc.py +17 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/voc/yolov5_s-v61_fast_1xb64-50e_voc.py +270 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/voc/yolov5_x-v61_fast_1xb32-50e_voc.py +26 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_l-p6-v62_syncbn_fast_8xb16-300e_coco.py +15 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_l-v61_syncbn_fast_8xb16-300e_coco.py +15 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco.py +79 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_m-v61_syncbn_fast_8xb16-300e_coco.py +79 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_n-p6-v62_syncbn_fast_8xb16-300e_coco.py +15 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_n-v61_syncbn_fast_8xb16-300e_coco.py +15 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco.py +138 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_s-v61_fast_1xb12-40e_608x352_cat.py +70 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_s-v61_fast_1xb12-40e_cat.py +56 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_s-v61_syncbn-detect_8xb16-300e_coco.py +23 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_s-v61_syncbn_8xb16-300e_coco.py +292 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_s-v61_syncbn_fast_1xb4-300e_balloon.py +42 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_s-v61_syncbn_fast_8xb16-300e_coco.py +12 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_x-p6-v62_syncbn_fast_8xb16-300e_coco.py +14 -0
- Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_x-v61_syncbn_fast_8xb16-300e_coco.py +14 -0
.idea/.gitignore
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Default ignored files
|
| 2 |
+
/shelf/
|
| 3 |
+
/workspace.xml
|
| 4 |
+
# Editor-based HTTP Client requests
|
| 5 |
+
/httpRequests/
|
| 6 |
+
# Datasource local storage ignored files
|
| 7 |
+
/dataSources/
|
| 8 |
+
/dataSources.local.xml
|
.idea/Forecast4Muses.iml
ADDED
|
@@ -0,0 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
<?xml version="1.0" encoding="UTF-8"?>
|
| 2 |
+
<module type="PYTHON_MODULE" version="4">
|
| 3 |
+
<component name="NewModuleRootManager">
|
| 4 |
+
<content url="file://$MODULE_DIR$">
|
| 5 |
+
<excludeFolder url="file://$MODULE_DIR$/venv" />
|
| 6 |
+
</content>
|
| 7 |
+
<orderEntry type="inheritedJdk" />
|
| 8 |
+
<orderEntry type="sourceFolder" forTests="false" />
|
| 9 |
+
</component>
|
| 10 |
+
<component name="PyDocumentationSettings">
|
| 11 |
+
<option name="format" value="GOOGLE" />
|
| 12 |
+
<option name="myDocStringFormat" value="Google" />
|
| 13 |
+
</component>
|
| 14 |
+
</module>
|
.idea/deployment.xml
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
<?xml version="1.0" encoding="UTF-8"?>
|
| 2 |
+
<project version="4">
|
| 3 |
+
<component name="PublishConfigData" remoteFilesAllowedToDisappearOnAutoupload="false">
|
| 4 |
+
<serverData>
|
| 5 |
+
<paths name="Forecast4Muses">
|
| 6 |
+
<serverdata>
|
| 7 |
+
<mappings>
|
| 8 |
+
<mapping local="$PROJECT_DIR$" web="/" />
|
| 9 |
+
</mappings>
|
| 10 |
+
</serverdata>
|
| 11 |
+
</paths>
|
| 12 |
+
<paths name="root@connect.beijinga.seetacloud.com:21383 password">
|
| 13 |
+
<serverdata>
|
| 14 |
+
<mappings>
|
| 15 |
+
<mapping local="$PROJECT_DIR$" web="/" />
|
| 16 |
+
</mappings>
|
| 17 |
+
</serverdata>
|
| 18 |
+
</paths>
|
| 19 |
+
</serverData>
|
| 20 |
+
</component>
|
| 21 |
+
</project>
|
.idea/inspectionProfiles/Project_Default.xml
ADDED
|
@@ -0,0 +1,75 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
<component name="InspectionProjectProfileManager">
|
| 2 |
+
<profile version="1.0">
|
| 3 |
+
<option name="myName" value="Project Default" />
|
| 4 |
+
<inspection_tool class="DuplicatedCode" enabled="true" level="WEAK WARNING" enabled_by_default="true">
|
| 5 |
+
<Languages>
|
| 6 |
+
<language minSize="74" name="Python" />
|
| 7 |
+
</Languages>
|
| 8 |
+
</inspection_tool>
|
| 9 |
+
<inspection_tool class="HtmlUnknownAttribute" enabled="true" level="WARNING" enabled_by_default="true">
|
| 10 |
+
<option name="myValues">
|
| 11 |
+
<value>
|
| 12 |
+
<list size="1">
|
| 13 |
+
<item index="0" class="java.lang.String" itemvalue="length" />
|
| 14 |
+
</list>
|
| 15 |
+
</value>
|
| 16 |
+
</option>
|
| 17 |
+
<option name="myCustomValuesEnabled" value="true" />
|
| 18 |
+
</inspection_tool>
|
| 19 |
+
<inspection_tool class="HtmlUnknownTag" enabled="true" level="WARNING" enabled_by_default="true">
|
| 20 |
+
<option name="myValues">
|
| 21 |
+
<value>
|
| 22 |
+
<list size="7">
|
| 23 |
+
<item index="0" class="java.lang.String" itemvalue="nobr" />
|
| 24 |
+
<item index="1" class="java.lang.String" itemvalue="noembed" />
|
| 25 |
+
<item index="2" class="java.lang.String" itemvalue="comment" />
|
| 26 |
+
<item index="3" class="java.lang.String" itemvalue="noscript" />
|
| 27 |
+
<item index="4" class="java.lang.String" itemvalue="embed" />
|
| 28 |
+
<item index="5" class="java.lang.String" itemvalue="script" />
|
| 29 |
+
<item index="6" class="java.lang.String" itemvalue="li" />
|
| 30 |
+
</list>
|
| 31 |
+
</value>
|
| 32 |
+
</option>
|
| 33 |
+
<option name="myCustomValuesEnabled" value="true" />
|
| 34 |
+
</inspection_tool>
|
| 35 |
+
<inspection_tool class="PyPackageRequirementsInspection" enabled="true" level="WARNING" enabled_by_default="true">
|
| 36 |
+
<option name="ignoredPackages">
|
| 37 |
+
<value>
|
| 38 |
+
<list size="15">
|
| 39 |
+
<item index="0" class="java.lang.String" itemvalue="json_tricks" />
|
| 40 |
+
<item index="1" class="java.lang.String" itemvalue="coverage" />
|
| 41 |
+
<item index="2" class="java.lang.String" itemvalue="scipy" />
|
| 42 |
+
<item index="3" class="java.lang.String" itemvalue="pytest" />
|
| 43 |
+
<item index="4" class="java.lang.String" itemvalue="chumpy" />
|
| 44 |
+
<item index="5" class="java.lang.String" itemvalue="flake8" />
|
| 45 |
+
<item index="6" class="java.lang.String" itemvalue="pytest-runner" />
|
| 46 |
+
<item index="7" class="java.lang.String" itemvalue="interrogate" />
|
| 47 |
+
<item index="8" class="java.lang.String" itemvalue="munkres" />
|
| 48 |
+
<item index="9" class="java.lang.String" itemvalue="parameterized" />
|
| 49 |
+
<item index="10" class="java.lang.String" itemvalue="xtcocotools" />
|
| 50 |
+
<item index="11" class="java.lang.String" itemvalue="isort" />
|
| 51 |
+
<item index="12" class="java.lang.String" itemvalue="xdoctest" />
|
| 52 |
+
<item index="13" class="java.lang.String" itemvalue="pandas" />
|
| 53 |
+
<item index="14" class="java.lang.String" itemvalue="xlwings" />
|
| 54 |
+
</list>
|
| 55 |
+
</value>
|
| 56 |
+
</option>
|
| 57 |
+
</inspection_tool>
|
| 58 |
+
<inspection_tool class="PyPep8NamingInspection" enabled="true" level="WEAK WARNING" enabled_by_default="true">
|
| 59 |
+
<option name="ignoredErrors">
|
| 60 |
+
<list>
|
| 61 |
+
<option value="N806" />
|
| 62 |
+
<option value="N802" />
|
| 63 |
+
<option value="N803" />
|
| 64 |
+
</list>
|
| 65 |
+
</option>
|
| 66 |
+
</inspection_tool>
|
| 67 |
+
<inspection_tool class="PyUnresolvedReferencesInspection" enabled="true" level="WARNING" enabled_by_default="true">
|
| 68 |
+
<option name="ignoredIdentifiers">
|
| 69 |
+
<list>
|
| 70 |
+
<option value="IndexError.__add__" />
|
| 71 |
+
</list>
|
| 72 |
+
</option>
|
| 73 |
+
</inspection_tool>
|
| 74 |
+
</profile>
|
| 75 |
+
</component>
|
.idea/inspectionProfiles/profiles_settings.xml
ADDED
|
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
<component name="InspectionProjectProfileManager">
|
| 2 |
+
<settings>
|
| 3 |
+
<option name="USE_PROJECT_PROFILE" value="false" />
|
| 4 |
+
<version value="1.0" />
|
| 5 |
+
</settings>
|
| 6 |
+
</component>
|
.idea/misc.xml
ADDED
|
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
<?xml version="1.0" encoding="UTF-8"?>
|
| 2 |
+
<project version="4">
|
| 3 |
+
<component name="ProjectRootManager" version="2" project-jdk-name="Python 3.8 (Forecast4Muses)" project-jdk-type="Python SDK" />
|
| 4 |
+
</project>
|
.idea/modules.xml
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
<?xml version="1.0" encoding="UTF-8"?>
|
| 2 |
+
<project version="4">
|
| 3 |
+
<component name="ProjectModuleManager">
|
| 4 |
+
<modules>
|
| 5 |
+
<module fileurl="file://$PROJECT_DIR$/.idea/Forecast4Muses.iml" filepath="$PROJECT_DIR$/.idea/Forecast4Muses.iml" />
|
| 6 |
+
</modules>
|
| 7 |
+
</component>
|
| 8 |
+
</project>
|
.idea/vcs.xml
ADDED
|
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
<?xml version="1.0" encoding="UTF-8"?>
|
| 2 |
+
<project version="4">
|
| 3 |
+
<component name="VcsDirectoryMappings">
|
| 4 |
+
<mapping directory="" vcs="Git" />
|
| 5 |
+
</component>
|
| 6 |
+
</project>
|
Model/Model6/Model6_0_ClothesDetection/__init__.py
ADDED
|
File without changes
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/__init__.py
ADDED
|
File without changes
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/_base_/__init__.py
ADDED
|
File without changes
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/_base_/default_runtime.py
ADDED
|
@@ -0,0 +1,34 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
default_scope = 'mmyolo'
|
| 2 |
+
|
| 3 |
+
default_hooks = dict(
|
| 4 |
+
timer=dict(type='IterTimerHook'),
|
| 5 |
+
logger=dict(type='LoggerHook', interval=50),
|
| 6 |
+
param_scheduler=dict(type='ParamSchedulerHook'),
|
| 7 |
+
checkpoint=dict(type='CheckpointHook', interval=1),
|
| 8 |
+
sampler_seed=dict(type='DistSamplerSeedHook'),
|
| 9 |
+
visualization=dict(type='mmdet.DetVisualizationHook'))
|
| 10 |
+
|
| 11 |
+
env_cfg = dict(
|
| 12 |
+
cudnn_benchmark=False,
|
| 13 |
+
mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0),
|
| 14 |
+
dist_cfg=dict(backend='nccl'),
|
| 15 |
+
)
|
| 16 |
+
|
| 17 |
+
vis_backends = [dict(type='LocalVisBackend')]
|
| 18 |
+
visualizer = dict(
|
| 19 |
+
type='mmdet.DetLocalVisualizer',
|
| 20 |
+
vis_backends=vis_backends,
|
| 21 |
+
name='visualizer')
|
| 22 |
+
log_processor = dict(type='LogProcessor', window_size=50, by_epoch=True)
|
| 23 |
+
|
| 24 |
+
log_level = 'INFO'
|
| 25 |
+
load_from = None
|
| 26 |
+
resume = False
|
| 27 |
+
|
| 28 |
+
# file_client_args = dict(
|
| 29 |
+
# backend='petrel',
|
| 30 |
+
# path_mapping=dict({
|
| 31 |
+
# './data/': 's3://openmmlab/datasets/detection/',
|
| 32 |
+
# 'data/': 's3://openmmlab/datasets/detection/'
|
| 33 |
+
# }))
|
| 34 |
+
file_client_args = dict(backend='disk')
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/_base_/det_p5_tta.py
ADDED
|
@@ -0,0 +1,57 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# TODO: Need to solve the problem of multiple file_client_args parameters
|
| 2 |
+
# _file_client_args = dict(
|
| 3 |
+
# backend='petrel',
|
| 4 |
+
# path_mapping=dict({
|
| 5 |
+
# './data/': 's3://openmmlab/datasets/detection/',
|
| 6 |
+
# 'data/': 's3://openmmlab/datasets/detection/'
|
| 7 |
+
# }))
|
| 8 |
+
_file_client_args = dict(backend='disk')
|
| 9 |
+
|
| 10 |
+
tta_model = dict(
|
| 11 |
+
type='mmdet.DetTTAModel',
|
| 12 |
+
tta_cfg=dict(nms=dict(type='nms', iou_threshold=0.65), max_per_img=300))
|
| 13 |
+
|
| 14 |
+
img_scales = [(640, 640), (320, 320), (960, 960)]
|
| 15 |
+
|
| 16 |
+
# LoadImageFromFile
|
| 17 |
+
# / | \
|
| 18 |
+
# (RatioResize,LetterResize) (RatioResize,LetterResize) (RatioResize,LetterResize) # noqa
|
| 19 |
+
# / \ / \ / \
|
| 20 |
+
# RandomFlip RandomFlip RandomFlip RandomFlip RandomFlip RandomFlip # noqa
|
| 21 |
+
# | | | | | |
|
| 22 |
+
# LoadAnn LoadAnn LoadAnn LoadAnn LoadAnn LoadAnn
|
| 23 |
+
# | | | | | |
|
| 24 |
+
# PackDetIn PackDetIn PackDetIn PackDetIn PackDetIn PackDetIn # noqa
|
| 25 |
+
|
| 26 |
+
_multiscale_resize_transforms = [
|
| 27 |
+
dict(
|
| 28 |
+
type='Compose',
|
| 29 |
+
transforms=[
|
| 30 |
+
dict(type='YOLOv5KeepRatioResize', scale=s),
|
| 31 |
+
dict(
|
| 32 |
+
type='LetterResize',
|
| 33 |
+
scale=s,
|
| 34 |
+
allow_scale_up=False,
|
| 35 |
+
pad_val=dict(img=114))
|
| 36 |
+
]) for s in img_scales
|
| 37 |
+
]
|
| 38 |
+
|
| 39 |
+
tta_pipeline = [
|
| 40 |
+
dict(type='LoadImageFromFile', file_client_args=_file_client_args),
|
| 41 |
+
dict(
|
| 42 |
+
type='TestTimeAug',
|
| 43 |
+
transforms=[
|
| 44 |
+
_multiscale_resize_transforms,
|
| 45 |
+
[
|
| 46 |
+
dict(type='mmdet.RandomFlip', prob=1.),
|
| 47 |
+
dict(type='mmdet.RandomFlip', prob=0.)
|
| 48 |
+
], [dict(type='mmdet.LoadAnnotations', with_bbox=True)],
|
| 49 |
+
[
|
| 50 |
+
dict(
|
| 51 |
+
type='mmdet.PackDetInputs',
|
| 52 |
+
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
|
| 53 |
+
'scale_factor', 'pad_param', 'flip',
|
| 54 |
+
'flip_direction'))
|
| 55 |
+
]
|
| 56 |
+
])
|
| 57 |
+
]
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/custom_dataset/.ipynb_checkpoints/__init__.py
ADDED
|
File without changes
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/custom_dataset/.ipynb_checkpoints/yolov5_s-v61_syncbn_fast_1xb32-100e_cat-checkpoint.py
ADDED
|
@@ -0,0 +1,135 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = '../yolov5/yolov5_s-v61_syncbn_fast_8xb16-300e_coco.py'
|
| 2 |
+
|
| 3 |
+
max_epochs = 100 # 训练的最大 epoch
|
| 4 |
+
data_root = './data-df2/' # 数据集目录的绝对路径
|
| 5 |
+
# data_root = '/root/workspace/mmyolo/data/cat/' # Docker 容器里面数据集目录的绝对路径
|
| 6 |
+
|
| 7 |
+
# 结果保存的路径，可以省略，省略保存的文件名位于 work_dirs 下 config 同名的文件夹中
|
| 8 |
+
# 如果某个 config 只是修改了部分参数，修改这个变量就可以将新的训练文件保存到其他地方
|
| 9 |
+
work_dir = './work_dirs/yolov5_s_df2'
|
| 10 |
+
|
| 11 |
+
# load_from 可以指定本地路径或者 URL，设置了 URL 会自动进行下载，因为上面已经下载过，我们这里设置本地路径
|
| 12 |
+
# 因为本教程是在 cat 数据集上微调，故这里需要使用 `load_from` 来加载 MMYOLO 中的预训练模型，这样可以在加快收敛速度的同时保证精度
|
| 13 |
+
# load_from = './work_dirs/yolov5_s-v61_syncbn_fast_8xb16-300e_coco_20220918_084700-86e02187.pth' # noqa
|
| 14 |
+
|
| 15 |
+
# 根据自己的 GPU 情况，修改 batch size，YOLOv5-s 默认为 8卡 x 16bs
|
| 16 |
+
train_batch_size_per_gpu = 32
|
| 17 |
+
train_num_workers = 4 # 推荐使用 train_num_workers = nGPU x 4
|
| 18 |
+
|
| 19 |
+
save_epoch_intervals = 2 # 每 interval 轮迭代进行一次保存一次权重
|
| 20 |
+
|
| 21 |
+
# 根据自己的 GPU 情况，修改 base_lr，修改的比例是 base_lr_default * (your_bs / default_bs)
|
| 22 |
+
base_lr = _base_.base_lr / 4
|
| 23 |
+
|
| 24 |
+
anchors = [ # 此处已经根据数据集特点更新了 anchor，关于 anchor 的生成，后面小节会讲解
|
| 25 |
+
[(68, 69), (154, 91), (143, 162)], # P3/8
|
| 26 |
+
[(242, 160), (189, 287), (391, 207)], # P4/16
|
| 27 |
+
[(353, 337), (539, 341), (443, 432)] # P5/32
|
| 28 |
+
]
|
| 29 |
+
|
| 30 |
+
class_name = ('short_sleeved_shirt',
|
| 31 |
+
'long_sleeved_shirt',
|
| 32 |
+
'short_sleeved_outwear',
|
| 33 |
+
'long_sleeved_outwear',
|
| 34 |
+
'vest',
|
| 35 |
+
'sling',
|
| 36 |
+
'shorts',
|
| 37 |
+
'trousers',
|
| 38 |
+
'skirt',
|
| 39 |
+
'short_sleeved_dress',
|
| 40 |
+
'long_sleeved_dress',
|
| 41 |
+
'vest_dress',
|
| 42 |
+
'sling_dress') # 根据 class_with_id.txt 类别信息，设置 class_name
|
| 43 |
+
|
| 44 |
+
num_classes = len(class_name)
|
| 45 |
+
metainfo = dict(
|
| 46 |
+
classes=class_name,
|
| 47 |
+
palette=[(255, 0, 0),
|
| 48 |
+
(255, 128, 0),
|
| 49 |
+
(255, 255, 0),
|
| 50 |
+
(128, 255, 0),
|
| 51 |
+
(0, 255, 0),
|
| 52 |
+
(0, 255, 128),
|
| 53 |
+
(0, 255, 255),
|
| 54 |
+
(0, 128, 255),
|
| 55 |
+
(0, 0, 255),
|
| 56 |
+
(127, 0, 255),
|
| 57 |
+
(255, 0, 255),
|
| 58 |
+
(255, 0, 127),
|
| 59 |
+
(128, 128, 128)] # 画图时候的颜色，随便设置即可
|
| 60 |
+
)
|
| 61 |
+
|
| 62 |
+
train_cfg = dict(
|
| 63 |
+
max_epochs=max_epochs,
|
| 64 |
+
val_begin=20, # 第几个 epoch 后验证，这里设置 20 是因为前 20 个 epoch 精度不高，测试意义不大，故跳过
|
| 65 |
+
val_interval=save_epoch_intervals # 每 val_interval 轮迭代进行一次测试评估
|
| 66 |
+
# dynamic_intervals=[(max_epochs-_base_.num_last_epochs, 1)]
|
| 67 |
+
)
|
| 68 |
+
|
| 69 |
+
model = dict(
|
| 70 |
+
bbox_head=dict(
|
| 71 |
+
head_module=dict(num_classes=num_classes),
|
| 72 |
+
prior_generator=dict(base_sizes=anchors),
|
| 73 |
+
|
| 74 |
+
# loss_cls 会根据 num_classes 动态调整，但是 num_classes = 1 的时候，loss_cls 恒为 0
|
| 75 |
+
loss_cls=dict(loss_weight=0.5 *
|
| 76 |
+
(num_classes / 80 * 3 / _base_.num_det_layers))))
|
| 77 |
+
|
| 78 |
+
train_dataloader = dict(
|
| 79 |
+
batch_size=train_batch_size_per_gpu,
|
| 80 |
+
num_workers=train_num_workers,
|
| 81 |
+
dataset=dict(
|
| 82 |
+
_delete_=True,
|
| 83 |
+
type='RepeatDataset',
|
| 84 |
+
# 数据量太少的话，可以使用 RepeatDataset ，在每个 epoch 内重复当前数据集 n 次，这里设置 5 是重复 5 次
|
| 85 |
+
times=2,
|
| 86 |
+
dataset=dict(
|
| 87 |
+
type=_base_.dataset_type,
|
| 88 |
+
data_root=data_root,
|
| 89 |
+
metainfo=metainfo,
|
| 90 |
+
ann_file='annotations/trainval.json',
|
| 91 |
+
data_prefix=dict(img='smaller-dataset/'),
|
| 92 |
+
filter_cfg=dict(filter_empty_gt=False, min_size=32),
|
| 93 |
+
pipeline=_base_.train_pipeline)))
|
| 94 |
+
|
| 95 |
+
val_dataloader = dict(
|
| 96 |
+
dataset=dict(
|
| 97 |
+
metainfo=metainfo,
|
| 98 |
+
data_root=data_root,
|
| 99 |
+
ann_file='annotations/trainval.json',
|
| 100 |
+
data_prefix=dict(img='smaller-dataset/')))
|
| 101 |
+
|
| 102 |
+
test_dataloader = val_dataloader
|
| 103 |
+
|
| 104 |
+
val_evaluator = dict(ann_file=data_root + 'annotations/trainval.json')
|
| 105 |
+
test_evaluator = val_evaluator
|
| 106 |
+
|
| 107 |
+
optim_wrapper = dict(optimizer=dict(lr=base_lr))
|
| 108 |
+
|
| 109 |
+
default_hooks = dict(
|
| 110 |
+
# 设置间隔多少个 epoch 保存模型，以及保存模型最多几个，`save_best` 是另外保存最佳模型（推荐）
|
| 111 |
+
checkpoint=dict(
|
| 112 |
+
type='CheckpointHook',
|
| 113 |
+
interval=save_epoch_intervals,
|
| 114 |
+
max_keep_ckpts=5,
|
| 115 |
+
save_best='auto'),
|
| 116 |
+
param_scheduler=dict(max_epochs=max_epochs, warmup_mim_iter=10),
|
| 117 |
+
# logger 输出的间隔
|
| 118 |
+
logger=dict(type='LoggerHook', interval=10))
|
| 119 |
+
|
| 120 |
+
# custom_hooks = [
|
| 121 |
+
# dict(
|
| 122 |
+
# type="EMAHook",
|
| 123 |
+
# ema_type="ExpMomentumEMA",
|
| 124 |
+
# momentum=0.0001,
|
| 125 |
+
# update_buffers=True,
|
| 126 |
+
# strict_load=False,
|
| 127 |
+
# priority=49),
|
| 128 |
+
# dict(
|
| 129 |
+
# type="mmdet.PipelineSwitchHook",
|
| 130 |
+
# switch_epoch=max_epochs-max_epochs-_base_.num_last_epochs,
|
| 131 |
+
# switch_pipeline=_base_.train_pipeline_stage2
|
| 132 |
+
# )
|
| 133 |
+
# ]
|
| 134 |
+
|
| 135 |
+
visualizer = dict(vis_backends=[dict(type='LocalVisBackend'), dict(type='WandbVisBackend')])
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/custom_dataset/.ipynb_checkpoints/yolov6_s_fast-checkpoint.py
ADDED
|
@@ -0,0 +1,124 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = '../yolov6/yolov6_s_syncbn_fast_8xb32-400e_coco.py'
|
| 2 |
+
|
| 3 |
+
max_epochs = 100 # 训练的最大 epoch
|
| 4 |
+
data_root = './data-df2/' # 数据集目录的绝对路径
|
| 5 |
+
|
| 6 |
+
# 结果保存的路径，可以省略，省略保存的文件名位于 work_dirs 下 config 同名的文件夹中
|
| 7 |
+
# 如果某个 config 只是修改了部分参数，修改这个变量就可以将新的训练文件保存到其他地方
|
| 8 |
+
work_dir = './work_dirs/yolov6_s_df2'
|
| 9 |
+
|
| 10 |
+
# 根据自己的 GPU 情况，修改 batch size，YOLOv5-s 默认为 8卡 x 16bs
|
| 11 |
+
train_batch_size_per_gpu = 32
|
| 12 |
+
train_num_workers = 4 # 推荐使用 train_num_workers = nGPU x 4
|
| 13 |
+
|
| 14 |
+
save_epoch_intervals = 2 # 每 interval 轮迭代进行一次保存一次权重
|
| 15 |
+
|
| 16 |
+
# 根据自己的 GPU 情况，修改 base_lr，修改的比例是 base_lr_default * (your_bs / default_bs)
|
| 17 |
+
base_lr = _base_.base_lr / 4
|
| 18 |
+
|
| 19 |
+
class_name = ('short_sleeved_shirt',
|
| 20 |
+
'long_sleeved_shirt',
|
| 21 |
+
'short_sleeved_outwear',
|
| 22 |
+
'long_sleeved_outwear',
|
| 23 |
+
'vest',
|
| 24 |
+
'sling',
|
| 25 |
+
'shorts',
|
| 26 |
+
'trousers',
|
| 27 |
+
'skirt',
|
| 28 |
+
'short_sleeved_dress',
|
| 29 |
+
'long_sleeved_dress',
|
| 30 |
+
'vest_dress',
|
| 31 |
+
'sling_dress') # 根据 class_with_id.txt 类别信息，设置 class_name
|
| 32 |
+
|
| 33 |
+
num_classes = len(class_name)
|
| 34 |
+
metainfo = dict(
|
| 35 |
+
classes=class_name,
|
| 36 |
+
palette=[(255, 0, 0),
|
| 37 |
+
(255, 128, 0),
|
| 38 |
+
(255, 255, 0),
|
| 39 |
+
(128, 255, 0),
|
| 40 |
+
(0, 255, 0),
|
| 41 |
+
(0, 255, 128),
|
| 42 |
+
(0, 255, 255),
|
| 43 |
+
(0, 128, 255),
|
| 44 |
+
(0, 0, 255),
|
| 45 |
+
(127, 0, 255),
|
| 46 |
+
(255, 0, 255),
|
| 47 |
+
(255, 0, 127),
|
| 48 |
+
(128, 128, 128)] # 画图时候的颜色，随便设置即可
|
| 49 |
+
)
|
| 50 |
+
|
| 51 |
+
train_cfg = dict(
|
| 52 |
+
max_epochs=max_epochs,
|
| 53 |
+
val_begin=20, # 第几个 epoch 后验证，这里设置 20 是因为前 20 个 epoch 精度不高，测试意义不大，故跳过
|
| 54 |
+
val_interval=save_epoch_intervals, # 每 val_interval 轮迭代进行一次测试评估
|
| 55 |
+
dynamic_intervals=[(max_epochs-_base_.num_last_epochs, 1)]
|
| 56 |
+
)
|
| 57 |
+
|
| 58 |
+
model = dict(
|
| 59 |
+
bbox_head=dict(
|
| 60 |
+
head_module=dict(num_classes=num_classes)),
|
| 61 |
+
train_cfg=dict(
|
| 62 |
+
initial_assigner=dict(num_classes=num_classes),
|
| 63 |
+
assigner=dict(num_classes=num_classes)
|
| 64 |
+
)
|
| 65 |
+
)
|
| 66 |
+
|
| 67 |
+
train_dataloader = dict(
|
| 68 |
+
batch_size=train_batch_size_per_gpu,
|
| 69 |
+
num_workers=train_num_workers,
|
| 70 |
+
dataset=dict(
|
| 71 |
+
_delete_=True,
|
| 72 |
+
type='RepeatDataset',
|
| 73 |
+
# 数据量太少的话，可以使用 RepeatDataset ，在每个 epoch 内重复当前数据集 n 次，这里设置 5 是重复 5 次
|
| 74 |
+
times=2,
|
| 75 |
+
dataset=dict(
|
| 76 |
+
type=_base_.dataset_type,
|
| 77 |
+
data_root=data_root,
|
| 78 |
+
metainfo=metainfo,
|
| 79 |
+
ann_file='annotations/trainval.json',
|
| 80 |
+
data_prefix=dict(img='smaller-dataset/'),
|
| 81 |
+
filter_cfg=dict(filter_empty_gt=False, min_size=32),
|
| 82 |
+
pipeline=_base_.train_pipeline)))
|
| 83 |
+
|
| 84 |
+
val_dataloader = dict(
|
| 85 |
+
dataset=dict(
|
| 86 |
+
metainfo=metainfo,
|
| 87 |
+
data_root=data_root,
|
| 88 |
+
ann_file='annotations/trainval.json',
|
| 89 |
+
data_prefix=dict(img='smaller-dataset/')))
|
| 90 |
+
|
| 91 |
+
test_dataloader = val_dataloader
|
| 92 |
+
|
| 93 |
+
val_evaluator = dict(ann_file=data_root + 'annotations/trainval.json')
|
| 94 |
+
test_evaluator = val_evaluator
|
| 95 |
+
|
| 96 |
+
optim_wrapper = dict(optimizer=dict(lr=base_lr))
|
| 97 |
+
|
| 98 |
+
default_hooks = dict(
|
| 99 |
+
# 设置间隔多少个 epoch 保存模型，以及保存模型最多几个，`save_best` 是另外保存最佳模型（推荐）
|
| 100 |
+
checkpoint=dict(
|
| 101 |
+
type='CheckpointHook',
|
| 102 |
+
interval=save_epoch_intervals,
|
| 103 |
+
max_keep_ckpts=5,
|
| 104 |
+
save_best='auto'),
|
| 105 |
+
param_scheduler=dict(max_epochs=max_epochs),
|
| 106 |
+
# logger 输出的间隔
|
| 107 |
+
logger=dict(type='LoggerHook', interval=10))
|
| 108 |
+
|
| 109 |
+
custom_hooks = [
|
| 110 |
+
dict(
|
| 111 |
+
type="EMAHook",
|
| 112 |
+
ema_type="ExpMomentumEMA",
|
| 113 |
+
momentum=0.0001,
|
| 114 |
+
update_buffers=True,
|
| 115 |
+
strict_load=False,
|
| 116 |
+
priority=49),
|
| 117 |
+
dict(
|
| 118 |
+
type="mmdet.PipelineSwitchHook",
|
| 119 |
+
switch_epoch=max_epochs-max_epochs-_base_.num_last_epochs,
|
| 120 |
+
switch_pipeline=_base_.train_pipeline_stage2
|
| 121 |
+
)
|
| 122 |
+
]
|
| 123 |
+
|
| 124 |
+
visualizer = dict(vis_backends=[dict(type='LocalVisBackend'), dict(type='WandbVisBackend')])
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/custom_dataset/[A]dataset_split.sh
ADDED
|
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
cd ..
|
| 2 |
+
python tools/misc/coco_split.py --json data-df2/deepfashion2-smaller-dataset.json \
|
| 3 |
+
--out-dir data-df2/annotations/ \
|
| 4 |
+
--ratios 8 2 \
|
| 5 |
+
--shuffle --seed 10
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/custom_dataset/__init__.py
ADDED
|
File without changes
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/custom_dataset/yolov5_s-v61_syncbn_fast_1xb32-100e_cat.py
ADDED
|
@@ -0,0 +1,135 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = '../yolov5/yolov5_s-v61_syncbn_fast_8xb16-300e_coco.py'
|
| 2 |
+
|
| 3 |
+
max_epochs = 100 # 训练的最大 epoch
|
| 4 |
+
data_root = './data-df2/' # 数据集目录的绝对路径
|
| 5 |
+
# data_root = '/root/workspace/mmyolo/data/cat/' # Docker 容器里面数据集目录的绝对路径
|
| 6 |
+
|
| 7 |
+
# 结果保存的路径，可以省略，省略保存的文件名位于 work_dirs 下 config 同名的文件夹中
|
| 8 |
+
# 如果某个 config 只是修改了部分参数，修改这个变量就可以将新的训练文件保存到其他地方
|
| 9 |
+
work_dir = './work_dirs/yolov5_s_df2'
|
| 10 |
+
|
| 11 |
+
# load_from 可以指定本地路径或者 URL，设置了 URL 会自动进行下载，因为上面已经下载过，我们这里设置本地路径
|
| 12 |
+
# 因为本教程是在 cat 数据集上微调，故这里需要使用 `load_from` 来加载 MMYOLO 中的预训练模型，这样可以在加快收敛速度的同时保证精度
|
| 13 |
+
# load_from = './work_dirs/yolov5_s-v61_syncbn_fast_8xb16-300e_coco_20220918_084700-86e02187.pth' # noqa
|
| 14 |
+
|
| 15 |
+
# 根据自己的 GPU 情况，修改 batch size，YOLOv5-s 默认为 8卡 x 16bs
|
| 16 |
+
train_batch_size_per_gpu = 32
|
| 17 |
+
train_num_workers = 4 # 推荐使用 train_num_workers = nGPU x 4
|
| 18 |
+
|
| 19 |
+
save_epoch_intervals = 2 # 每 interval 轮迭代进行一次保存一次权重
|
| 20 |
+
|
| 21 |
+
# 根据自己的 GPU 情况，修改 base_lr，修改的比例是 base_lr_default * (your_bs / default_bs)
|
| 22 |
+
base_lr = _base_.base_lr / 4
|
| 23 |
+
|
| 24 |
+
anchors = [ # 此处已经根据数据集特点更新了 anchor，关于 anchor 的生成，后面小节会讲解
|
| 25 |
+
[(68, 69), (154, 91), (143, 162)], # P3/8
|
| 26 |
+
[(242, 160), (189, 287), (391, 207)], # P4/16
|
| 27 |
+
[(353, 337), (539, 341), (443, 432)] # P5/32
|
| 28 |
+
]
|
| 29 |
+
|
| 30 |
+
class_name = ('short_sleeved_shirt',
|
| 31 |
+
'long_sleeved_shirt',
|
| 32 |
+
'short_sleeved_outwear',
|
| 33 |
+
'long_sleeved_outwear',
|
| 34 |
+
'vest',
|
| 35 |
+
'sling',
|
| 36 |
+
'shorts',
|
| 37 |
+
'trousers',
|
| 38 |
+
'skirt',
|
| 39 |
+
'short_sleeved_dress',
|
| 40 |
+
'long_sleeved_dress',
|
| 41 |
+
'vest_dress',
|
| 42 |
+
'sling_dress') # 根据 class_with_id.txt 类别信息，设置 class_name
|
| 43 |
+
|
| 44 |
+
num_classes = len(class_name)
|
| 45 |
+
metainfo = dict(
|
| 46 |
+
classes=class_name,
|
| 47 |
+
palette=[(255, 0, 0),
|
| 48 |
+
(255, 128, 0),
|
| 49 |
+
(255, 255, 0),
|
| 50 |
+
(128, 255, 0),
|
| 51 |
+
(0, 255, 0),
|
| 52 |
+
(0, 255, 128),
|
| 53 |
+
(0, 255, 255),
|
| 54 |
+
(0, 128, 255),
|
| 55 |
+
(0, 0, 255),
|
| 56 |
+
(127, 0, 255),
|
| 57 |
+
(255, 0, 255),
|
| 58 |
+
(255, 0, 127),
|
| 59 |
+
(128, 128, 128)] # 画图时候的颜色，随便设置即可
|
| 60 |
+
)
|
| 61 |
+
|
| 62 |
+
train_cfg = dict(
|
| 63 |
+
max_epochs=max_epochs,
|
| 64 |
+
val_begin=20, # 第几个 epoch 后验证，这里设置 20 是因为前 20 个 epoch 精度不高，测试意义不大，故跳过
|
| 65 |
+
val_interval=save_epoch_intervals # 每 val_interval 轮迭代进行一次测试评估
|
| 66 |
+
# dynamic_intervals=[(max_epochs-_base_.num_last_epochs, 1)]
|
| 67 |
+
)
|
| 68 |
+
|
| 69 |
+
model = dict(
|
| 70 |
+
bbox_head=dict(
|
| 71 |
+
head_module=dict(num_classes=num_classes),
|
| 72 |
+
prior_generator=dict(base_sizes=anchors),
|
| 73 |
+
|
| 74 |
+
# loss_cls 会根据 num_classes 动态调整，但是 num_classes = 1 的时候，loss_cls 恒为 0
|
| 75 |
+
loss_cls=dict(loss_weight=0.5 *
|
| 76 |
+
(num_classes / 80 * 3 / _base_.num_det_layers))))
|
| 77 |
+
|
| 78 |
+
train_dataloader = dict(
|
| 79 |
+
batch_size=train_batch_size_per_gpu,
|
| 80 |
+
num_workers=train_num_workers,
|
| 81 |
+
dataset=dict(
|
| 82 |
+
_delete_=True,
|
| 83 |
+
type='RepeatDataset',
|
| 84 |
+
# 数据量太少的话，可以使用 RepeatDataset ，在每个 epoch 内重复当前数据集 n 次，这里设置 5 是重复 5 次
|
| 85 |
+
times=2,
|
| 86 |
+
dataset=dict(
|
| 87 |
+
type=_base_.dataset_type,
|
| 88 |
+
data_root=data_root,
|
| 89 |
+
metainfo=metainfo,
|
| 90 |
+
ann_file='annotations/trainval.json',
|
| 91 |
+
data_prefix=dict(img='smaller-dataset/'),
|
| 92 |
+
filter_cfg=dict(filter_empty_gt=False, min_size=32),
|
| 93 |
+
pipeline=_base_.train_pipeline)))
|
| 94 |
+
|
| 95 |
+
val_dataloader = dict(
|
| 96 |
+
dataset=dict(
|
| 97 |
+
metainfo=metainfo,
|
| 98 |
+
data_root=data_root,
|
| 99 |
+
ann_file='annotations/trainval.json',
|
| 100 |
+
data_prefix=dict(img='smaller-dataset/')))
|
| 101 |
+
|
| 102 |
+
test_dataloader = val_dataloader
|
| 103 |
+
|
| 104 |
+
val_evaluator = dict(ann_file=data_root + 'annotations/trainval.json')
|
| 105 |
+
test_evaluator = val_evaluator
|
| 106 |
+
|
| 107 |
+
optim_wrapper = dict(optimizer=dict(lr=base_lr))
|
| 108 |
+
|
| 109 |
+
default_hooks = dict(
|
| 110 |
+
# 设置间隔多少个 epoch 保存模型，以及保存模型最多几个，`save_best` 是另外保存最佳模型（推荐）
|
| 111 |
+
checkpoint=dict(
|
| 112 |
+
type='CheckpointHook',
|
| 113 |
+
interval=save_epoch_intervals,
|
| 114 |
+
max_keep_ckpts=5,
|
| 115 |
+
save_best='auto'),
|
| 116 |
+
param_scheduler=dict(max_epochs=max_epochs, warmup_mim_iter=10),
|
| 117 |
+
# logger 输出的间隔
|
| 118 |
+
logger=dict(type='LoggerHook', interval=10))
|
| 119 |
+
|
| 120 |
+
# custom_hooks = [
|
| 121 |
+
# dict(
|
| 122 |
+
# type="EMAHook",
|
| 123 |
+
# ema_type="ExpMomentumEMA",
|
| 124 |
+
# momentum=0.0001,
|
| 125 |
+
# update_buffers=True,
|
| 126 |
+
# strict_load=False,
|
| 127 |
+
# priority=49),
|
| 128 |
+
# dict(
|
| 129 |
+
# type="mmdet.PipelineSwitchHook",
|
| 130 |
+
# switch_epoch=max_epochs-max_epochs-_base_.num_last_epochs,
|
| 131 |
+
# switch_pipeline=_base_.train_pipeline_stage2
|
| 132 |
+
# )
|
| 133 |
+
# ]
|
| 134 |
+
|
| 135 |
+
visualizer = dict(vis_backends=[dict(type='LocalVisBackend'), dict(type='WandbVisBackend')])
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/custom_dataset/yolov6_s_fast.py
ADDED
|
@@ -0,0 +1,124 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = '../yolov6/yolov6_s_syncbn_fast_8xb32-400e_coco.py'
|
| 2 |
+
|
| 3 |
+
max_epochs = 100 # 训练的最大 epoch
|
| 4 |
+
data_root = './data-df2/' # 数据集目录的绝对路径
|
| 5 |
+
|
| 6 |
+
# 结果保存的路径，可以省略，省略保存的文件名位于 work_dirs 下 config 同名的文件夹中
|
| 7 |
+
# 如果某个 config 只是修改了部分参数，修改这个变量就可以将新的训练文件保存到其他地方
|
| 8 |
+
work_dir = './work_dirs/yolov6_s_df2'
|
| 9 |
+
|
| 10 |
+
# 根据自己的 GPU 情况，修改 batch size，YOLOv5-s 默认为 8卡 x 16bs
|
| 11 |
+
train_batch_size_per_gpu = 32
|
| 12 |
+
train_num_workers = 4 # 推荐使用 train_num_workers = nGPU x 4
|
| 13 |
+
|
| 14 |
+
save_epoch_intervals = 2
|
| 15 |
+
|
| 16 |
+
# 根据自己的 GPU 情况，修改 base_lr，修改的比例是 base_lr_default * (your_bs / default_bs)
|
| 17 |
+
base_lr = _base_.base_lr / 4
|
| 18 |
+
|
| 19 |
+
class_name = ('short_sleeved_shirt',
|
| 20 |
+
'long_sleeved_shirt',
|
| 21 |
+
'short_sleeved_outwear',
|
| 22 |
+
'long_sleeved_outwear',
|
| 23 |
+
'vest',
|
| 24 |
+
'sling',
|
| 25 |
+
'shorts',
|
| 26 |
+
'trousers',
|
| 27 |
+
'skirt',
|
| 28 |
+
'short_sleeved_dress',
|
| 29 |
+
'long_sleeved_dress',
|
| 30 |
+
'vest_dress',
|
| 31 |
+
'sling_dress') # 根据 class_with_id.txt 类别信息，设置 class_name
|
| 32 |
+
|
| 33 |
+
num_classes = len(class_name)
|
| 34 |
+
metainfo = dict(
|
| 35 |
+
classes=class_name,
|
| 36 |
+
palette=[(255, 0, 0),
|
| 37 |
+
(255, 128, 0),
|
| 38 |
+
(255, 255, 0),
|
| 39 |
+
(128, 255, 0),
|
| 40 |
+
(0, 255, 0),
|
| 41 |
+
(0, 255, 128),
|
| 42 |
+
(0, 255, 255),
|
| 43 |
+
(0, 128, 255),
|
| 44 |
+
(0, 0, 255),
|
| 45 |
+
(127, 0, 255),
|
| 46 |
+
(255, 0, 255),
|
| 47 |
+
(255, 0, 127),
|
| 48 |
+
(128, 128, 128)] # 画图时候的颜色，随便设置即可
|
| 49 |
+
)
|
| 50 |
+
|
| 51 |
+
train_cfg = dict(
|
| 52 |
+
max_epochs=max_epochs,
|
| 53 |
+
val_begin=20, # 第几个 epoch 后验证，这里设置 20 是因为前 20 个 epoch 精度不高，测试意义不大，故跳过
|
| 54 |
+
val_interval=save_epoch_intervals, # 每 val_interval 轮迭代进行一次测试评估
|
| 55 |
+
dynamic_intervals=[(max_epochs-_base_.num_last_epochs, 1)]
|
| 56 |
+
)
|
| 57 |
+
|
| 58 |
+
model = dict(
|
| 59 |
+
bbox_head=dict(
|
| 60 |
+
head_module=dict(num_classes=num_classes)),
|
| 61 |
+
train_cfg=dict(
|
| 62 |
+
initial_assigner=dict(num_classes=num_classes),
|
| 63 |
+
assigner=dict(num_classes=num_classes)
|
| 64 |
+
)
|
| 65 |
+
)
|
| 66 |
+
|
| 67 |
+
train_dataloader = dict(
|
| 68 |
+
batch_size=train_batch_size_per_gpu,
|
| 69 |
+
num_workers=train_num_workers,
|
| 70 |
+
dataset=dict(
|
| 71 |
+
_delete_=True,
|
| 72 |
+
type='RepeatDataset',
|
| 73 |
+
# 数据量太少的话，可以使用 RepeatDataset ，在每个 epoch 内重复当前数据集 n 次，这里设置 5 是重复 5 次
|
| 74 |
+
times=2,
|
| 75 |
+
dataset=dict(
|
| 76 |
+
type=_base_.dataset_type,
|
| 77 |
+
data_root=data_root,
|
| 78 |
+
metainfo=metainfo,
|
| 79 |
+
ann_file='annotations/trainval.json',
|
| 80 |
+
data_prefix=dict(img='smaller-dataset/'),
|
| 81 |
+
filter_cfg=dict(filter_empty_gt=False, min_size=32),
|
| 82 |
+
pipeline=_base_.train_pipeline)))
|
| 83 |
+
|
| 84 |
+
val_dataloader = dict(
|
| 85 |
+
dataset=dict(
|
| 86 |
+
metainfo=metainfo,
|
| 87 |
+
data_root=data_root,
|
| 88 |
+
ann_file='annotations/trainval.json',
|
| 89 |
+
data_prefix=dict(img='smaller-dataset/')))
|
| 90 |
+
|
| 91 |
+
test_dataloader = val_dataloader
|
| 92 |
+
|
| 93 |
+
val_evaluator = dict(ann_file=data_root + 'annotations/trainval.json')
|
| 94 |
+
test_evaluator = val_evaluator
|
| 95 |
+
|
| 96 |
+
optim_wrapper = dict(optimizer=dict(lr=base_lr))
|
| 97 |
+
|
| 98 |
+
default_hooks = dict(
|
| 99 |
+
# 设置间隔多少个 epoch 保存模型，以及保存模型最多几个，`save_best` 是另外保存最佳模型（推荐）
|
| 100 |
+
checkpoint=dict(
|
| 101 |
+
type='CheckpointHook',
|
| 102 |
+
interval=save_epoch_intervals,
|
| 103 |
+
max_keep_ckpts=5,
|
| 104 |
+
save_best='auto'),
|
| 105 |
+
param_scheduler=dict(max_epochs=max_epochs),
|
| 106 |
+
# logger 输出的间隔
|
| 107 |
+
logger=dict(type='LoggerHook', interval=10))
|
| 108 |
+
|
| 109 |
+
custom_hooks = [
|
| 110 |
+
dict(
|
| 111 |
+
type="EMAHook",
|
| 112 |
+
ema_type="ExpMomentumEMA",
|
| 113 |
+
momentum=0.0001,
|
| 114 |
+
update_buffers=True,
|
| 115 |
+
strict_load=False,
|
| 116 |
+
priority=49),
|
| 117 |
+
dict(
|
| 118 |
+
type="mmdet.PipelineSwitchHook",
|
| 119 |
+
switch_epoch=max_epochs-max_epochs-_base_.num_last_epochs,
|
| 120 |
+
switch_pipeline=_base_.train_pipeline_stage2
|
| 121 |
+
)
|
| 122 |
+
]
|
| 123 |
+
|
| 124 |
+
visualizer = dict(vis_backends=[dict(type='LocalVisBackend'), dict(type='WandbVisBackend')])
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/custom_dataset/yolov7_l_syncbn_fast_6x16b-100e_coco.py
ADDED
|
@@ -0,0 +1,489 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = ['../_base_/default_runtime.py', '../_base_/det_p5_tta.py']
|
| 2 |
+
|
| 3 |
+
data_root = './data-df2/'
|
| 4 |
+
train_ann_file = 'annotations/train.json'
|
| 5 |
+
train_data_prefix = 'smaller-dataset/'
|
| 6 |
+
val_ann_file = 'annotations/val.json'
|
| 7 |
+
val_data_prefix = 'smaller-dataset/'
|
| 8 |
+
test_ann_file = 'annotations/test.json'
|
| 9 |
+
test_data_prefix = 'smaller-dataset/'
|
| 10 |
+
# num_classes = 13
|
| 11 |
+
train_batch_size_per_gpu = 32
|
| 12 |
+
train_num_workers = 4
|
| 13 |
+
persistent_workers = True
|
| 14 |
+
|
| 15 |
+
vis_backends = [
|
| 16 |
+
dict(type='LocalVisBackend'),
|
| 17 |
+
]
|
| 18 |
+
visualizer = dict(
|
| 19 |
+
type='mmdet.DetLocalVisualizer',
|
| 20 |
+
vis_backends=[
|
| 21 |
+
dict(type='LocalVisBackend'),
|
| 22 |
+
# dict(type='WandbVisBackend'),
|
| 23 |
+
dict(type='TensorboardVisBackend')
|
| 24 |
+
],
|
| 25 |
+
name='visualizer')
|
| 26 |
+
log_processor = dict(type='LogProcessor', window_size=50, by_epoch=True)
|
| 27 |
+
log_level = 'INFO'
|
| 28 |
+
load_from = None
|
| 29 |
+
resume = False
|
| 30 |
+
|
| 31 |
+
anchors = [
|
| 32 |
+
[(12, 16), (19, 36), (40, 28)], # P3/8
|
| 33 |
+
[(36, 75), (76, 55), (72, 146)], # P4/16
|
| 34 |
+
[(142, 110), (192, 243), (459, 401)] # P5/32
|
| 35 |
+
]
|
| 36 |
+
|
| 37 |
+
base_lr = 0.01
|
| 38 |
+
max_epochs = 100
|
| 39 |
+
|
| 40 |
+
num_epoch_stage2 = 10 # The last 10 epochs switch evaluation interval
|
| 41 |
+
val_interval_stage2 = 1
|
| 42 |
+
|
| 43 |
+
model_test_cfg = dict(
|
| 44 |
+
multi_label=True,
|
| 45 |
+
nms_pre=30000,
|
| 46 |
+
score_thr=0.001,
|
| 47 |
+
nms=dict(type='nms', iou_threshold=0.65),
|
| 48 |
+
max_per_img=300)
|
| 49 |
+
|
| 50 |
+
img_scale = (640, 640)
|
| 51 |
+
dataset_type = 'YOLOv5CocoDataset'
|
| 52 |
+
classes=('short_sleeved_shirt', 'long_sleeved_shirt',
|
| 53 |
+
'short_sleeved_outwear', 'long_sleeved_outwear',
|
| 54 |
+
'vest', 'sling', 'shorts', 'trousers', 'skirt',
|
| 55 |
+
'short_sleeved_dress', 'long_sleeved_dress',
|
| 56 |
+
'vest_dress', 'sling_dress')
|
| 57 |
+
num_classes = len(classes)
|
| 58 |
+
palette=[(255, 0, 0), (255, 128, 0), (255, 255, 0),
|
| 59 |
+
(128, 255, 0), (0, 255, 0), (0, 255, 128),
|
| 60 |
+
(0, 255, 255), (0, 128, 255), (0, 0, 255),
|
| 61 |
+
(127, 0, 255), (255, 0, 255), (255, 0, 127),
|
| 62 |
+
(128, 128, 128)]
|
| 63 |
+
metainfo = dict(
|
| 64 |
+
classes=classes,
|
| 65 |
+
palette=palette
|
| 66 |
+
)
|
| 67 |
+
val_batch_size_per_gpu = 1
|
| 68 |
+
val_num_workers = 2
|
| 69 |
+
batch_shapes_cfg = dict(
|
| 70 |
+
type='BatchShapePolicy',
|
| 71 |
+
batch_size=val_batch_size_per_gpu,
|
| 72 |
+
img_size=img_scale[0],
|
| 73 |
+
size_divisor=32,
|
| 74 |
+
extra_pad_ratio=0.5)
|
| 75 |
+
strides = [8, 16, 32] # Strides of multi-scale prior box
|
| 76 |
+
num_det_layers = 3
|
| 77 |
+
norm_cfg = dict(type='BN', momentum=0.03, eps=0.001)
|
| 78 |
+
|
| 79 |
+
# Data augmentation
|
| 80 |
+
max_translate_ratio = 0.2 # YOLOv5RandomAffine
|
| 81 |
+
scaling_ratio_range = (0.1, 2.0) # YOLOv5RandomAffine
|
| 82 |
+
mixup_prob = 0.15 # YOLOv5MixUp
|
| 83 |
+
randchoice_mosaic_prob = [0.8, 0.2]
|
| 84 |
+
mixup_alpha = 8.0 # YOLOv5MixUp
|
| 85 |
+
mixup_beta = 8.0 # YOLOv5MixUp
|
| 86 |
+
|
| 87 |
+
# -----train val related-----
|
| 88 |
+
loss_cls_weight = 0.3
|
| 89 |
+
loss_bbox_weight = 0.05
|
| 90 |
+
loss_obj_weight = 0.7
|
| 91 |
+
# BatchYOLOv7Assigner params
|
| 92 |
+
simota_candidate_topk = 10
|
| 93 |
+
simota_iou_weight = 3.0
|
| 94 |
+
simota_cls_weight = 1.0
|
| 95 |
+
prior_match_thr = 4. # Priori box matching threshold
|
| 96 |
+
obj_level_weights = [4., 1.,
|
| 97 |
+
0.4] # The obj loss weights of the three output layers
|
| 98 |
+
|
| 99 |
+
lr_factor = 0.1 # Learning rate scaling factor
|
| 100 |
+
weight_decay = 0.0005
|
| 101 |
+
save_epoch_intervals = 1
|
| 102 |
+
max_keep_ckpts = 5
|
| 103 |
+
|
| 104 |
+
env_cfg = dict(
|
| 105 |
+
cudnn_benchmark=True,
|
| 106 |
+
mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0),
|
| 107 |
+
dist_cfg=dict(backend='nccl'))
|
| 108 |
+
|
| 109 |
+
# ===============================Unmodified in most cases====================
|
| 110 |
+
model = dict(
|
| 111 |
+
type='YOLODetector',
|
| 112 |
+
data_preprocessor=dict(
|
| 113 |
+
type='YOLOv5DetDataPreprocessor',
|
| 114 |
+
mean=[0., 0., 0.],
|
| 115 |
+
std=[255., 255., 255.],
|
| 116 |
+
bgr_to_rgb=True),
|
| 117 |
+
backbone=dict(
|
| 118 |
+
type='YOLOv7Backbone',
|
| 119 |
+
arch='L',
|
| 120 |
+
norm_cfg=norm_cfg,
|
| 121 |
+
act_cfg=dict(type='SiLU', inplace=True)),
|
| 122 |
+
neck=dict(
|
| 123 |
+
type='YOLOv7PAFPN',
|
| 124 |
+
block_cfg=dict(
|
| 125 |
+
type='ELANBlock',
|
| 126 |
+
middle_ratio=0.5,
|
| 127 |
+
block_ratio=0.25,
|
| 128 |
+
num_blocks=4,
|
| 129 |
+
num_convs_in_block=1),
|
| 130 |
+
upsample_feats_cat_first=False,
|
| 131 |
+
in_channels=[512, 1024, 1024],
|
| 132 |
+
# The real output channel will be multiplied by 2
|
| 133 |
+
out_channels=[128, 256, 512],
|
| 134 |
+
norm_cfg=norm_cfg,
|
| 135 |
+
act_cfg=dict(type='SiLU', inplace=True)),
|
| 136 |
+
bbox_head=dict(
|
| 137 |
+
type='YOLOv7Head',
|
| 138 |
+
head_module=dict(
|
| 139 |
+
type='YOLOv7HeadModule',
|
| 140 |
+
num_classes=num_classes,
|
| 141 |
+
in_channels=[256, 512, 1024],
|
| 142 |
+
featmap_strides=strides,
|
| 143 |
+
num_base_priors=3),
|
| 144 |
+
prior_generator=dict(
|
| 145 |
+
type='mmdet.YOLOAnchorGenerator',
|
| 146 |
+
base_sizes=anchors,
|
| 147 |
+
strides=strides),
|
| 148 |
+
# scaled based on number of detection layers
|
| 149 |
+
loss_cls=dict(
|
| 150 |
+
type='mmdet.CrossEntropyLoss',
|
| 151 |
+
use_sigmoid=True,
|
| 152 |
+
reduction='mean',
|
| 153 |
+
loss_weight=loss_cls_weight *
|
| 154 |
+
(num_classes / 80 * 3 / num_det_layers)),
|
| 155 |
+
loss_bbox=dict(
|
| 156 |
+
type='IoULoss',
|
| 157 |
+
iou_mode='ciou',
|
| 158 |
+
bbox_format='xyxy',
|
| 159 |
+
reduction='mean',
|
| 160 |
+
loss_weight=loss_bbox_weight * (3 / num_det_layers),
|
| 161 |
+
return_iou=True),
|
| 162 |
+
loss_obj=dict(
|
| 163 |
+
type='mmdet.CrossEntropyLoss',
|
| 164 |
+
use_sigmoid=True,
|
| 165 |
+
reduction='mean',
|
| 166 |
+
loss_weight=loss_obj_weight *
|
| 167 |
+
((img_scale[0] / 640)**2 * 3 / num_det_layers)),
|
| 168 |
+
prior_match_thr=prior_match_thr,
|
| 169 |
+
obj_level_weights=obj_level_weights,
|
| 170 |
+
# BatchYOLOv7Assigner params
|
| 171 |
+
simota_candidate_topk=simota_candidate_topk,
|
| 172 |
+
simota_iou_weight=simota_iou_weight,
|
| 173 |
+
simota_cls_weight=simota_cls_weight),
|
| 174 |
+
test_cfg=model_test_cfg)
|
| 175 |
+
|
| 176 |
+
pre_transform = [
|
| 177 |
+
dict(type='LoadImageFromFile', file_client_args=_base_.file_client_args),
|
| 178 |
+
dict(type='LoadAnnotations', with_bbox=True)
|
| 179 |
+
]
|
| 180 |
+
|
| 181 |
+
mosiac4_pipeline = [
|
| 182 |
+
dict(
|
| 183 |
+
type='Mosaic',
|
| 184 |
+
img_scale=img_scale,
|
| 185 |
+
pad_val=114.0,
|
| 186 |
+
pre_transform=pre_transform),
|
| 187 |
+
dict(
|
| 188 |
+
type='YOLOv5RandomAffine',
|
| 189 |
+
max_rotate_degree=0.0,
|
| 190 |
+
max_shear_degree=0.0,
|
| 191 |
+
max_translate_ratio=max_translate_ratio, # note
|
| 192 |
+
scaling_ratio_range=scaling_ratio_range, # note
|
| 193 |
+
# img_scale is (width, height)
|
| 194 |
+
border=(-img_scale[0] // 2, -img_scale[1] // 2),
|
| 195 |
+
border_val=(114, 114, 114)),
|
| 196 |
+
]
|
| 197 |
+
|
| 198 |
+
mosiac9_pipeline = [
|
| 199 |
+
dict(
|
| 200 |
+
type='Mosaic9',
|
| 201 |
+
img_scale=img_scale,
|
| 202 |
+
pad_val=114.0,
|
| 203 |
+
pre_transform=pre_transform),
|
| 204 |
+
dict(
|
| 205 |
+
type='YOLOv5RandomAffine',
|
| 206 |
+
max_rotate_degree=0.0,
|
| 207 |
+
max_shear_degree=0.0,
|
| 208 |
+
max_translate_ratio=max_translate_ratio, # note
|
| 209 |
+
scaling_ratio_range=scaling_ratio_range, # note
|
| 210 |
+
# img_scale is (width, height)
|
| 211 |
+
border=(-img_scale[0] // 2, -img_scale[1] // 2),
|
| 212 |
+
border_val=(114, 114, 114)),
|
| 213 |
+
]
|
| 214 |
+
|
| 215 |
+
randchoice_mosaic_pipeline = dict(
|
| 216 |
+
type='RandomChoice',
|
| 217 |
+
transforms=[mosiac4_pipeline, mosiac9_pipeline],
|
| 218 |
+
prob=randchoice_mosaic_prob)
|
| 219 |
+
|
| 220 |
+
train_pipeline = [
|
| 221 |
+
*pre_transform,
|
| 222 |
+
randchoice_mosaic_pipeline,
|
| 223 |
+
dict(
|
| 224 |
+
type='YOLOv5MixUp',
|
| 225 |
+
alpha=mixup_alpha, # note
|
| 226 |
+
beta=mixup_beta, # note
|
| 227 |
+
prob=mixup_prob,
|
| 228 |
+
pre_transform=[*pre_transform, randchoice_mosaic_pipeline]),
|
| 229 |
+
dict(type='YOLOv5HSVRandomAug'),
|
| 230 |
+
dict(type='mmdet.RandomFlip', prob=0.5),
|
| 231 |
+
dict(
|
| 232 |
+
type='mmdet.PackDetInputs',
|
| 233 |
+
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape', 'flip',
|
| 234 |
+
'flip_direction'))
|
| 235 |
+
]
|
| 236 |
+
|
| 237 |
+
test_pipeline = [
|
| 238 |
+
dict(type='LoadImageFromFile', file_client_args=_base_.file_client_args),
|
| 239 |
+
dict(type='YOLOv5KeepRatioResize', scale=img_scale),
|
| 240 |
+
dict(
|
| 241 |
+
type='LetterResize',
|
| 242 |
+
scale=img_scale,
|
| 243 |
+
allow_scale_up=False,
|
| 244 |
+
pad_val=dict(img=114)),
|
| 245 |
+
dict(type='LoadAnnotations', with_bbox=True, _scope_='mmdet'),
|
| 246 |
+
dict(
|
| 247 |
+
type='mmdet.PackDetInputs',
|
| 248 |
+
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
|
| 249 |
+
'scale_factor', 'pad_param'))
|
| 250 |
+
]
|
| 251 |
+
|
| 252 |
+
train_dataloader = dict(
|
| 253 |
+
batch_size=train_batch_size_per_gpu,
|
| 254 |
+
num_workers=train_num_workers,
|
| 255 |
+
persistent_workers=persistent_workers,
|
| 256 |
+
pin_memory=True,
|
| 257 |
+
sampler=dict(type='DefaultSampler', shuffle=True),
|
| 258 |
+
collate_fn=dict(type='yolov5_collate'), # FASTER
|
| 259 |
+
dataset=dict(
|
| 260 |
+
type='RepeatDataset',
|
| 261 |
+
times=2,
|
| 262 |
+
dataset=dict(
|
| 263 |
+
type=dataset_type,
|
| 264 |
+
data_root=data_root,
|
| 265 |
+
metainfo=metainfo,
|
| 266 |
+
ann_file=val_ann_file,
|
| 267 |
+
data_prefix=dict(img=train_data_prefix),
|
| 268 |
+
filter_cfg=dict(filter_empty_gt=False, min_size=32),
|
| 269 |
+
pipeline=train_pipeline)
|
| 270 |
+
)
|
| 271 |
+
)
|
| 272 |
+
|
| 273 |
+
val_dataloader = dict(
|
| 274 |
+
dataset=dict(
|
| 275 |
+
metainfo=metainfo,
|
| 276 |
+
data_root=data_root,
|
| 277 |
+
ann_file=val_ann_file,
|
| 278 |
+
data_prefix=dict(img=val_data_prefix)))
|
| 279 |
+
|
| 280 |
+
val_evaluator = dict(ann_file=data_root + val_ann_file)
|
| 281 |
+
|
| 282 |
+
test_dataloader = dict(
|
| 283 |
+
dataset=dict(
|
| 284 |
+
metainfo=metainfo,
|
| 285 |
+
data_root=data_root,
|
| 286 |
+
ann_file=test_ann_file,
|
| 287 |
+
data_prefix=dict(img=test_data_prefix)))
|
| 288 |
+
test_evaluator = dict(ann_file=data_root + test_ann_file)
|
| 289 |
+
|
| 290 |
+
train_cfg = dict(
|
| 291 |
+
type='EpochBasedTrainLoop',
|
| 292 |
+
max_epochs=max_epochs,
|
| 293 |
+
val_interval=save_epoch_intervals,
|
| 294 |
+
dynamic_intervals=[(max_epochs - num_epoch_stage2, val_interval_stage2)])
|
| 295 |
+
val_cfg = dict(type='ValLoop')
|
| 296 |
+
test_cfg = dict(type='TestLoop')
|
| 297 |
+
|
| 298 |
+
param_scheduler = None
|
| 299 |
+
optim_wrapper = dict(
|
| 300 |
+
type='OptimWrapper',
|
| 301 |
+
optimizer=dict(
|
| 302 |
+
type='SGD',
|
| 303 |
+
lr=base_lr,
|
| 304 |
+
momentum=0.937,
|
| 305 |
+
weight_decay=weight_decay,
|
| 306 |
+
nesterov=True,
|
| 307 |
+
batch_size_per_gpu=train_batch_size_per_gpu),
|
| 308 |
+
constructor='YOLOv7OptimWrapperConstructor')
|
| 309 |
+
|
| 310 |
+
# TO DO: change param_scheduler type to StepLR, refer to mobilenet
|
| 311 |
+
default_scope = 'mmyolo'
|
| 312 |
+
default_hooks = dict(
|
| 313 |
+
timer=dict(type='IterTimerHook'),
|
| 314 |
+
logger=dict(type='LoggerHook', interval=10),
|
| 315 |
+
param_scheduler=dict(
|
| 316 |
+
type='YOLOv5ParamSchedulerHook',
|
| 317 |
+
scheduler_type='cosine',
|
| 318 |
+
lr_factor=lr_factor, # note
|
| 319 |
+
max_epochs=max_epochs),
|
| 320 |
+
checkpoint=dict(
|
| 321 |
+
type='CheckpointHook',
|
| 322 |
+
save_param_scheduler=False,
|
| 323 |
+
interval=save_epoch_intervals,
|
| 324 |
+
save_best='auto',
|
| 325 |
+
max_keep_ckpts=max_keep_ckpts),
|
| 326 |
+
sampler_seed=dict(type='DistSamplerSeedHook'),
|
| 327 |
+
visualization=dict(type='mmdet.DetVisualizationHook'))
|
| 328 |
+
|
| 329 |
+
custom_hooks = [
|
| 330 |
+
dict(
|
| 331 |
+
type='EMAHook',
|
| 332 |
+
ema_type='ExpMomentumEMA',
|
| 333 |
+
momentum=0.001,
|
| 334 |
+
update_buffers=True,
|
| 335 |
+
strict_load=False,
|
| 336 |
+
priority=49)
|
| 337 |
+
]
|
| 338 |
+
|
| 339 |
+
# ============================
|
| 340 |
+
|
| 341 |
+
file_client_args = dict(backend='disk')
|
| 342 |
+
_file_client_args = dict(backend='disk')
|
| 343 |
+
tta_model = dict(
|
| 344 |
+
type='mmdet.DetTTAModel',
|
| 345 |
+
tta_cfg=dict(nms=dict(type='nms', iou_threshold=0.65), max_per_img=300))
|
| 346 |
+
img_scales = [
|
| 347 |
+
(
|
| 348 |
+
640,
|
| 349 |
+
640,
|
| 350 |
+
),
|
| 351 |
+
(
|
| 352 |
+
320,
|
| 353 |
+
320,
|
| 354 |
+
),
|
| 355 |
+
(
|
| 356 |
+
960,
|
| 357 |
+
960,
|
| 358 |
+
),
|
| 359 |
+
]
|
| 360 |
+
_multiscale_resize_transforms = [
|
| 361 |
+
dict(
|
| 362 |
+
type='Compose',
|
| 363 |
+
transforms=[
|
| 364 |
+
dict(type='YOLOv5KeepRatioResize', scale=(
|
| 365 |
+
640,
|
| 366 |
+
640,
|
| 367 |
+
)),
|
| 368 |
+
dict(
|
| 369 |
+
type='LetterResize',
|
| 370 |
+
scale=(
|
| 371 |
+
640,
|
| 372 |
+
640,
|
| 373 |
+
),
|
| 374 |
+
allow_scale_up=False,
|
| 375 |
+
pad_val=dict(img=114)),
|
| 376 |
+
]),
|
| 377 |
+
dict(
|
| 378 |
+
type='Compose',
|
| 379 |
+
transforms=[
|
| 380 |
+
dict(type='YOLOv5KeepRatioResize', scale=(
|
| 381 |
+
320,
|
| 382 |
+
320,
|
| 383 |
+
)),
|
| 384 |
+
dict(
|
| 385 |
+
type='LetterResize',
|
| 386 |
+
scale=(
|
| 387 |
+
320,
|
| 388 |
+
320,
|
| 389 |
+
),
|
| 390 |
+
allow_scale_up=False,
|
| 391 |
+
pad_val=dict(img=114)),
|
| 392 |
+
]),
|
| 393 |
+
dict(
|
| 394 |
+
type='Compose',
|
| 395 |
+
transforms=[
|
| 396 |
+
dict(type='YOLOv5KeepRatioResize', scale=(
|
| 397 |
+
960,
|
| 398 |
+
960,
|
| 399 |
+
)),
|
| 400 |
+
dict(
|
| 401 |
+
type='LetterResize',
|
| 402 |
+
scale=(
|
| 403 |
+
960,
|
| 404 |
+
960,
|
| 405 |
+
),
|
| 406 |
+
allow_scale_up=False,
|
| 407 |
+
pad_val=dict(img=114)),
|
| 408 |
+
]),
|
| 409 |
+
]
|
| 410 |
+
tta_pipeline = [
|
| 411 |
+
dict(type='LoadImageFromFile', file_client_args=dict(backend='disk')),
|
| 412 |
+
dict(
|
| 413 |
+
type='TestTimeAug',
|
| 414 |
+
transforms=[
|
| 415 |
+
[
|
| 416 |
+
dict(
|
| 417 |
+
type='Compose',
|
| 418 |
+
transforms=[
|
| 419 |
+
dict(type='YOLOv5KeepRatioResize', scale=(
|
| 420 |
+
640,
|
| 421 |
+
640,
|
| 422 |
+
)),
|
| 423 |
+
dict(
|
| 424 |
+
type='LetterResize',
|
| 425 |
+
scale=(
|
| 426 |
+
640,
|
| 427 |
+
640,
|
| 428 |
+
),
|
| 429 |
+
allow_scale_up=False,
|
| 430 |
+
pad_val=dict(img=114)),
|
| 431 |
+
]),
|
| 432 |
+
dict(
|
| 433 |
+
type='Compose',
|
| 434 |
+
transforms=[
|
| 435 |
+
dict(type='YOLOv5KeepRatioResize', scale=(
|
| 436 |
+
320,
|
| 437 |
+
320,
|
| 438 |
+
)),
|
| 439 |
+
dict(
|
| 440 |
+
type='LetterResize',
|
| 441 |
+
scale=(
|
| 442 |
+
320,
|
| 443 |
+
320,
|
| 444 |
+
),
|
| 445 |
+
allow_scale_up=False,
|
| 446 |
+
pad_val=dict(img=114)),
|
| 447 |
+
]),
|
| 448 |
+
dict(
|
| 449 |
+
type='Compose',
|
| 450 |
+
transforms=[
|
| 451 |
+
dict(type='YOLOv5KeepRatioResize', scale=(
|
| 452 |
+
960,
|
| 453 |
+
960,
|
| 454 |
+
)),
|
| 455 |
+
dict(
|
| 456 |
+
type='LetterResize',
|
| 457 |
+
scale=(
|
| 458 |
+
960,
|
| 459 |
+
960,
|
| 460 |
+
),
|
| 461 |
+
allow_scale_up=False,
|
| 462 |
+
pad_val=dict(img=114)),
|
| 463 |
+
]),
|
| 464 |
+
],
|
| 465 |
+
[
|
| 466 |
+
dict(type='mmdet.RandomFlip', prob=1.0),
|
| 467 |
+
dict(type='mmdet.RandomFlip', prob=0.0),
|
| 468 |
+
],
|
| 469 |
+
[
|
| 470 |
+
dict(type='mmdet.LoadAnnotations', with_bbox=True),
|
| 471 |
+
],
|
| 472 |
+
[
|
| 473 |
+
dict(
|
| 474 |
+
type='mmdet.PackDetInputs',
|
| 475 |
+
meta_keys=(
|
| 476 |
+
'img_id',
|
| 477 |
+
'img_path',
|
| 478 |
+
'ori_shape',
|
| 479 |
+
'img_shape',
|
| 480 |
+
'scale_factor',
|
| 481 |
+
'pad_param',
|
| 482 |
+
'flip',
|
| 483 |
+
'flip_direction',
|
| 484 |
+
)),
|
| 485 |
+
],
|
| 486 |
+
]),
|
| 487 |
+
]
|
| 488 |
+
|
| 489 |
+
launcher = 'none'
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/custom_dataset/yolov7_l_syncbn_fast_6x32b-100e_coco.py
ADDED
|
@@ -0,0 +1,554 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = ['../_base_/default_runtime.py', '../_base_/det_p5_tta.py']
|
| 2 |
+
|
| 3 |
+
data_root = './data-df2/'
|
| 4 |
+
train_ann_file = 'annotations/train.json'
|
| 5 |
+
train_data_prefix = 'smaller-dataset/'
|
| 6 |
+
val_ann_file = 'annotations/val.json'
|
| 7 |
+
val_data_prefix = 'smaller-dataset/'
|
| 8 |
+
test_ann_file = 'annotations/test.json'
|
| 9 |
+
test_data_prefix = 'smaller-dataset/'
|
| 10 |
+
num_classes = 13
|
| 11 |
+
train_batch_size_per_gpu = 32
|
| 12 |
+
train_num_workers = 24
|
| 13 |
+
persistent_workers = True
|
| 14 |
+
|
| 15 |
+
vis_backends = [
|
| 16 |
+
dict(type='LocalVisBackend'),
|
| 17 |
+
]
|
| 18 |
+
visualizer = dict(
|
| 19 |
+
type='mmdet.DetLocalVisualizer',
|
| 20 |
+
vis_backends=[
|
| 21 |
+
dict(type='LocalVisBackend'),
|
| 22 |
+
dict(type='WandbVisBackend'),
|
| 23 |
+
# dict(type='TensorboardVisBackend')
|
| 24 |
+
],
|
| 25 |
+
name='visualizer')
|
| 26 |
+
log_processor = dict(type='LogProcessor', window_size=50, by_epoch=True)
|
| 27 |
+
log_level = 'INFO'
|
| 28 |
+
load_from = None
|
| 29 |
+
resume = False
|
| 30 |
+
|
| 31 |
+
anchors = [
|
| 32 |
+
[(12, 16), (19, 36), (40, 28)], # P3/8
|
| 33 |
+
[(36, 75), (76, 55), (72, 146)], # P4/16
|
| 34 |
+
[(142, 110), (192, 243), (459, 401)] # P5/32
|
| 35 |
+
]
|
| 36 |
+
|
| 37 |
+
base_lr = 0.01
|
| 38 |
+
max_epochs = 100
|
| 39 |
+
|
| 40 |
+
num_epoch_stage2 = 10 # The last 10 epochs switch evaluation interval
|
| 41 |
+
val_interval_stage2 = 1
|
| 42 |
+
|
| 43 |
+
model_test_cfg = dict(
|
| 44 |
+
multi_label=True,
|
| 45 |
+
nms_pre=30000,
|
| 46 |
+
score_thr=0.001,
|
| 47 |
+
nms=dict(type='nms', iou_threshold=0.65),
|
| 48 |
+
max_per_img=300)
|
| 49 |
+
|
| 50 |
+
img_scale = (640, 640)
|
| 51 |
+
dataset_type = 'YOLOv5CocoDataset'
|
| 52 |
+
metainfo = dict(
|
| 53 |
+
classes=(
|
| 54 |
+
'short_sleeved_shirt', 'long_sleeved_shirt',
|
| 55 |
+
'short_sleeved_outwear', 'long_sleeved_outwear',
|
| 56 |
+
'vest', 'sling', 'shorts', 'trousers', 'skirt',
|
| 57 |
+
'short_sleeved_dress', 'long_sleeved_dress',
|
| 58 |
+
'vest_dress', 'sling_dress'),
|
| 59 |
+
palette=[
|
| 60 |
+
(255, 0, 0),
|
| 61 |
+
(
|
| 62 |
+
255,
|
| 63 |
+
128,
|
| 64 |
+
0,
|
| 65 |
+
),
|
| 66 |
+
(
|
| 67 |
+
255,
|
| 68 |
+
255,
|
| 69 |
+
0,
|
| 70 |
+
),
|
| 71 |
+
(
|
| 72 |
+
128,
|
| 73 |
+
255,
|
| 74 |
+
0,
|
| 75 |
+
),
|
| 76 |
+
(
|
| 77 |
+
0,
|
| 78 |
+
255,
|
| 79 |
+
0,
|
| 80 |
+
),
|
| 81 |
+
(
|
| 82 |
+
0,
|
| 83 |
+
255,
|
| 84 |
+
128,
|
| 85 |
+
),
|
| 86 |
+
(
|
| 87 |
+
0,
|
| 88 |
+
255,
|
| 89 |
+
255,
|
| 90 |
+
),
|
| 91 |
+
(
|
| 92 |
+
0,
|
| 93 |
+
128,
|
| 94 |
+
255,
|
| 95 |
+
),
|
| 96 |
+
(
|
| 97 |
+
0,
|
| 98 |
+
0,
|
| 99 |
+
255,
|
| 100 |
+
),
|
| 101 |
+
(
|
| 102 |
+
127,
|
| 103 |
+
0,
|
| 104 |
+
255,
|
| 105 |
+
),
|
| 106 |
+
(
|
| 107 |
+
255,
|
| 108 |
+
0,
|
| 109 |
+
255,
|
| 110 |
+
),
|
| 111 |
+
(
|
| 112 |
+
255,
|
| 113 |
+
0,
|
| 114 |
+
127,
|
| 115 |
+
),
|
| 116 |
+
(
|
| 117 |
+
128,
|
| 118 |
+
128,
|
| 119 |
+
128,
|
| 120 |
+
),
|
| 121 |
+
])
|
| 122 |
+
|
| 123 |
+
val_batch_size_per_gpu = 1
|
| 124 |
+
val_num_workers = 2
|
| 125 |
+
batch_shapes_cfg = dict(
|
| 126 |
+
type='BatchShapePolicy',
|
| 127 |
+
batch_size=val_batch_size_per_gpu,
|
| 128 |
+
img_size=img_scale[0],
|
| 129 |
+
size_divisor=32,
|
| 130 |
+
extra_pad_ratio=0.5)
|
| 131 |
+
strides = [8, 16, 32] # Strides of multi-scale prior box
|
| 132 |
+
num_det_layers = 3
|
| 133 |
+
norm_cfg = dict(type='BN', momentum=0.03, eps=0.001)
|
| 134 |
+
|
| 135 |
+
# Data augmentation
|
| 136 |
+
max_translate_ratio = 0.2 # YOLOv5RandomAffine
|
| 137 |
+
scaling_ratio_range = (0.1, 2.0) # YOLOv5RandomAffine
|
| 138 |
+
mixup_prob = 0.15 # YOLOv5MixUp
|
| 139 |
+
randchoice_mosaic_prob = [0.8, 0.2]
|
| 140 |
+
mixup_alpha = 8.0 # YOLOv5MixUp
|
| 141 |
+
mixup_beta = 8.0 # YOLOv5MixUp
|
| 142 |
+
|
| 143 |
+
# -----train val related-----
|
| 144 |
+
loss_cls_weight = 0.3
|
| 145 |
+
loss_bbox_weight = 0.05
|
| 146 |
+
loss_obj_weight = 0.7
|
| 147 |
+
# BatchYOLOv7Assigner params
|
| 148 |
+
simota_candidate_topk = 10
|
| 149 |
+
simota_iou_weight = 3.0
|
| 150 |
+
simota_cls_weight = 1.0
|
| 151 |
+
prior_match_thr = 4. # Priori box matching threshold
|
| 152 |
+
obj_level_weights = [4., 1.,
|
| 153 |
+
0.4] # The obj loss weights of the three output layers
|
| 154 |
+
|
| 155 |
+
lr_factor = 0.1 # Learning rate scaling factor
|
| 156 |
+
weight_decay = 0.0005
|
| 157 |
+
save_epoch_intervals = 1
|
| 158 |
+
max_keep_ckpts = 5
|
| 159 |
+
|
| 160 |
+
env_cfg = dict(
|
| 161 |
+
cudnn_benchmark=True,
|
| 162 |
+
mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0),
|
| 163 |
+
dist_cfg=dict(backend='nccl'))
|
| 164 |
+
|
| 165 |
+
# ===============================Unmodified in most cases====================
|
| 166 |
+
model = dict(
|
| 167 |
+
type='YOLODetector',
|
| 168 |
+
data_preprocessor=dict(
|
| 169 |
+
type='YOLOv5DetDataPreprocessor',
|
| 170 |
+
mean=[0., 0., 0.],
|
| 171 |
+
std=[255., 255., 255.],
|
| 172 |
+
bgr_to_rgb=True),
|
| 173 |
+
backbone=dict(
|
| 174 |
+
type='YOLOv7Backbone',
|
| 175 |
+
arch='L',
|
| 176 |
+
norm_cfg=norm_cfg,
|
| 177 |
+
act_cfg=dict(type='SiLU', inplace=True)),
|
| 178 |
+
neck=dict(
|
| 179 |
+
type='YOLOv7PAFPN',
|
| 180 |
+
block_cfg=dict(
|
| 181 |
+
type='ELANBlock',
|
| 182 |
+
middle_ratio=0.5,
|
| 183 |
+
block_ratio=0.25,
|
| 184 |
+
num_blocks=4,
|
| 185 |
+
num_convs_in_block=1),
|
| 186 |
+
upsample_feats_cat_first=False,
|
| 187 |
+
in_channels=[512, 1024, 1024],
|
| 188 |
+
# The real output channel will be multiplied by 2
|
| 189 |
+
out_channels=[128, 256, 512],
|
| 190 |
+
norm_cfg=norm_cfg,
|
| 191 |
+
act_cfg=dict(type='SiLU', inplace=True)),
|
| 192 |
+
bbox_head=dict(
|
| 193 |
+
type='YOLOv7Head',
|
| 194 |
+
head_module=dict(
|
| 195 |
+
type='YOLOv7HeadModule',
|
| 196 |
+
num_classes=num_classes,
|
| 197 |
+
in_channels=[256, 512, 1024],
|
| 198 |
+
featmap_strides=strides,
|
| 199 |
+
num_base_priors=3),
|
| 200 |
+
prior_generator=dict(
|
| 201 |
+
type='mmdet.YOLOAnchorGenerator',
|
| 202 |
+
base_sizes=anchors,
|
| 203 |
+
strides=strides),
|
| 204 |
+
# scaled based on number of detection layers
|
| 205 |
+
loss_cls=dict(
|
| 206 |
+
type='mmdet.CrossEntropyLoss',
|
| 207 |
+
use_sigmoid=True,
|
| 208 |
+
reduction='mean',
|
| 209 |
+
loss_weight=loss_cls_weight *
|
| 210 |
+
(num_classes / 80 * 3 / num_det_layers)),
|
| 211 |
+
loss_bbox=dict(
|
| 212 |
+
type='IoULoss',
|
| 213 |
+
iou_mode='ciou',
|
| 214 |
+
bbox_format='xyxy',
|
| 215 |
+
reduction='mean',
|
| 216 |
+
loss_weight=loss_bbox_weight * (3 / num_det_layers),
|
| 217 |
+
return_iou=True),
|
| 218 |
+
loss_obj=dict(
|
| 219 |
+
type='mmdet.CrossEntropyLoss',
|
| 220 |
+
use_sigmoid=True,
|
| 221 |
+
reduction='mean',
|
| 222 |
+
loss_weight=loss_obj_weight *
|
| 223 |
+
((img_scale[0] / 640)**2 * 3 / num_det_layers)),
|
| 224 |
+
prior_match_thr=prior_match_thr,
|
| 225 |
+
obj_level_weights=obj_level_weights,
|
| 226 |
+
# BatchYOLOv7Assigner params
|
| 227 |
+
simota_candidate_topk=simota_candidate_topk,
|
| 228 |
+
simota_iou_weight=simota_iou_weight,
|
| 229 |
+
simota_cls_weight=simota_cls_weight),
|
| 230 |
+
test_cfg=model_test_cfg)
|
| 231 |
+
|
| 232 |
+
pre_transform = [
|
| 233 |
+
dict(type='LoadImageFromFile', file_client_args=_base_.file_client_args),
|
| 234 |
+
dict(type='LoadAnnotations', with_bbox=True)
|
| 235 |
+
]
|
| 236 |
+
|
| 237 |
+
mosiac4_pipeline = [
|
| 238 |
+
dict(
|
| 239 |
+
type='Mosaic',
|
| 240 |
+
img_scale=img_scale,
|
| 241 |
+
pad_val=114.0,
|
| 242 |
+
pre_transform=pre_transform),
|
| 243 |
+
dict(
|
| 244 |
+
type='YOLOv5RandomAffine',
|
| 245 |
+
max_rotate_degree=0.0,
|
| 246 |
+
max_shear_degree=0.0,
|
| 247 |
+
max_translate_ratio=max_translate_ratio, # note
|
| 248 |
+
scaling_ratio_range=scaling_ratio_range, # note
|
| 249 |
+
# img_scale is (width, height)
|
| 250 |
+
border=(-img_scale[0] // 2, -img_scale[1] // 2),
|
| 251 |
+
border_val=(114, 114, 114)),
|
| 252 |
+
]
|
| 253 |
+
|
| 254 |
+
mosiac9_pipeline = [
|
| 255 |
+
dict(
|
| 256 |
+
type='Mosaic9',
|
| 257 |
+
img_scale=img_scale,
|
| 258 |
+
pad_val=114.0,
|
| 259 |
+
pre_transform=pre_transform),
|
| 260 |
+
dict(
|
| 261 |
+
type='YOLOv5RandomAffine',
|
| 262 |
+
max_rotate_degree=0.0,
|
| 263 |
+
max_shear_degree=0.0,
|
| 264 |
+
max_translate_ratio=max_translate_ratio, # note
|
| 265 |
+
scaling_ratio_range=scaling_ratio_range, # note
|
| 266 |
+
# img_scale is (width, height)
|
| 267 |
+
border=(-img_scale[0] // 2, -img_scale[1] // 2),
|
| 268 |
+
border_val=(114, 114, 114)),
|
| 269 |
+
]
|
| 270 |
+
|
| 271 |
+
randchoice_mosaic_pipeline = dict(
|
| 272 |
+
type='RandomChoice',
|
| 273 |
+
transforms=[mosiac4_pipeline, mosiac9_pipeline],
|
| 274 |
+
prob=randchoice_mosaic_prob)
|
| 275 |
+
|
| 276 |
+
train_pipeline = [
|
| 277 |
+
*pre_transform,
|
| 278 |
+
randchoice_mosaic_pipeline,
|
| 279 |
+
dict(
|
| 280 |
+
type='YOLOv5MixUp',
|
| 281 |
+
alpha=mixup_alpha, # note
|
| 282 |
+
beta=mixup_beta, # note
|
| 283 |
+
prob=mixup_prob,
|
| 284 |
+
pre_transform=[*pre_transform, randchoice_mosaic_pipeline]),
|
| 285 |
+
dict(type='YOLOv5HSVRandomAug'),
|
| 286 |
+
dict(type='mmdet.RandomFlip', prob=0.5),
|
| 287 |
+
dict(
|
| 288 |
+
type='mmdet.PackDetInputs',
|
| 289 |
+
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape', 'flip',
|
| 290 |
+
'flip_direction'))
|
| 291 |
+
]
|
| 292 |
+
|
| 293 |
+
test_pipeline = [
|
| 294 |
+
dict(type='LoadImageFromFile', file_client_args=_base_.file_client_args),
|
| 295 |
+
dict(type='YOLOv5KeepRatioResize', scale=img_scale),
|
| 296 |
+
dict(
|
| 297 |
+
type='LetterResize',
|
| 298 |
+
scale=img_scale,
|
| 299 |
+
allow_scale_up=False,
|
| 300 |
+
pad_val=dict(img=114)),
|
| 301 |
+
dict(type='LoadAnnotations', with_bbox=True, _scope_='mmdet'),
|
| 302 |
+
dict(
|
| 303 |
+
type='mmdet.PackDetInputs',
|
| 304 |
+
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
|
| 305 |
+
'scale_factor', 'pad_param'))
|
| 306 |
+
]
|
| 307 |
+
|
| 308 |
+
train_dataloader = dict(
|
| 309 |
+
batch_size=train_batch_size_per_gpu,
|
| 310 |
+
num_workers=train_num_workers,
|
| 311 |
+
persistent_workers=persistent_workers,
|
| 312 |
+
pin_memory=True,
|
| 313 |
+
sampler=dict(type='DefaultSampler', shuffle=True),
|
| 314 |
+
collate_fn=dict(type='yolov5_collate'), # FASTER
|
| 315 |
+
dataset=dict(
|
| 316 |
+
type='RepeatDataset',
|
| 317 |
+
times=2,
|
| 318 |
+
dataset=dict(
|
| 319 |
+
type=dataset_type,
|
| 320 |
+
metainfo=metainfo,
|
| 321 |
+
data_root=data_root,
|
| 322 |
+
ann_file=val_ann_file,
|
| 323 |
+
data_prefix=dict(img=train_data_prefix),
|
| 324 |
+
filter_cfg=dict(filter_empty_gt=False, min_size=32),
|
| 325 |
+
pipeline=train_pipeline
|
| 326 |
+
)
|
| 327 |
+
)
|
| 328 |
+
)
|
| 329 |
+
val_dataloader = dict(
|
| 330 |
+
dataset=dict(
|
| 331 |
+
type=dataset_type,
|
| 332 |
+
metainfo=metainfo,
|
| 333 |
+
data_root=data_root,
|
| 334 |
+
ann_file=val_ann_file,
|
| 335 |
+
data_prefix=dict(img=val_data_prefix)))
|
| 336 |
+
|
| 337 |
+
val_evaluator = dict(
|
| 338 |
+
type='mmdet.CocoMetric',
|
| 339 |
+
proposal_nums=(100, 1, 10),
|
| 340 |
+
ann_file=data_root + val_ann_file,
|
| 341 |
+
metric='bbox')
|
| 342 |
+
|
| 343 |
+
test_dataloader = dict(
|
| 344 |
+
dataset=dict(
|
| 345 |
+
metainfo=metainfo,
|
| 346 |
+
data_root=data_root,
|
| 347 |
+
ann_file=test_ann_file,
|
| 348 |
+
data_prefix=dict(img=test_data_prefix)))
|
| 349 |
+
test_evaluator = dict(
|
| 350 |
+
type='mmdet.CocoMetric',
|
| 351 |
+
proposal_nums=(100, 1, 10),
|
| 352 |
+
ann_file=data_root + test_ann_file,
|
| 353 |
+
metric='bbox')
|
| 354 |
+
|
| 355 |
+
train_cfg = dict(
|
| 356 |
+
type='EpochBasedTrainLoop',
|
| 357 |
+
max_epochs=max_epochs,
|
| 358 |
+
val_interval=save_epoch_intervals,
|
| 359 |
+
dynamic_intervals=[(max_epochs - num_epoch_stage2, val_interval_stage2)])
|
| 360 |
+
val_cfg = dict(type='ValLoop')
|
| 361 |
+
test_cfg = dict(type='TestLoop')
|
| 362 |
+
|
| 363 |
+
param_scheduler = None
|
| 364 |
+
optim_wrapper = dict(
|
| 365 |
+
type='OptimWrapper',
|
| 366 |
+
optimizer=dict(
|
| 367 |
+
type='SGD',
|
| 368 |
+
lr=base_lr,
|
| 369 |
+
momentum=0.937,
|
| 370 |
+
weight_decay=weight_decay,
|
| 371 |
+
nesterov=True,
|
| 372 |
+
batch_size_per_gpu=train_batch_size_per_gpu),
|
| 373 |
+
constructor='YOLOv7OptimWrapperConstructor')
|
| 374 |
+
|
| 375 |
+
# TO DO: change param_scheduler type to StepLR, refer to mobilenet
|
| 376 |
+
default_scope = 'mmyolo'
|
| 377 |
+
default_hooks = dict(
|
| 378 |
+
timer=dict(type='IterTimerHook'),
|
| 379 |
+
logger=dict(type='LoggerHook', interval=10),
|
| 380 |
+
param_scheduler=dict(
|
| 381 |
+
type='YOLOv5ParamSchedulerHook',
|
| 382 |
+
scheduler_type='cosine',
|
| 383 |
+
lr_factor=lr_factor, # note
|
| 384 |
+
max_epochs=max_epochs),
|
| 385 |
+
checkpoint=dict(
|
| 386 |
+
type='CheckpointHook',
|
| 387 |
+
save_param_scheduler=False,
|
| 388 |
+
interval=save_epoch_intervals,
|
| 389 |
+
save_best='auto',
|
| 390 |
+
max_keep_ckpts=max_keep_ckpts),
|
| 391 |
+
sampler_seed=dict(type='DistSamplerSeedHook'),
|
| 392 |
+
visualization=dict(type='mmdet.DetVisualizationHook'))
|
| 393 |
+
|
| 394 |
+
custom_hooks = [
|
| 395 |
+
dict(
|
| 396 |
+
type='EMAHook',
|
| 397 |
+
ema_type='ExpMomentumEMA',
|
| 398 |
+
momentum=0.001,
|
| 399 |
+
update_buffers=True,
|
| 400 |
+
strict_load=False,
|
| 401 |
+
priority=49)
|
| 402 |
+
]
|
| 403 |
+
|
| 404 |
+
# ============================
|
| 405 |
+
|
| 406 |
+
file_client_args = dict(backend='disk')
|
| 407 |
+
_file_client_args = dict(backend='disk')
|
| 408 |
+
tta_model = dict(
|
| 409 |
+
type='mmdet.DetTTAModel',
|
| 410 |
+
tta_cfg=dict(nms=dict(type='nms', iou_threshold=0.65), max_per_img=300))
|
| 411 |
+
img_scales = [
|
| 412 |
+
(
|
| 413 |
+
640,
|
| 414 |
+
640,
|
| 415 |
+
),
|
| 416 |
+
(
|
| 417 |
+
320,
|
| 418 |
+
320,
|
| 419 |
+
),
|
| 420 |
+
(
|
| 421 |
+
960,
|
| 422 |
+
960,
|
| 423 |
+
),
|
| 424 |
+
]
|
| 425 |
+
_multiscale_resize_transforms = [
|
| 426 |
+
dict(
|
| 427 |
+
type='Compose',
|
| 428 |
+
transforms=[
|
| 429 |
+
dict(type='YOLOv5KeepRatioResize', scale=(
|
| 430 |
+
640,
|
| 431 |
+
640,
|
| 432 |
+
)),
|
| 433 |
+
dict(
|
| 434 |
+
type='LetterResize',
|
| 435 |
+
scale=(
|
| 436 |
+
640,
|
| 437 |
+
640,
|
| 438 |
+
),
|
| 439 |
+
allow_scale_up=False,
|
| 440 |
+
pad_val=dict(img=114)),
|
| 441 |
+
]),
|
| 442 |
+
dict(
|
| 443 |
+
type='Compose',
|
| 444 |
+
transforms=[
|
| 445 |
+
dict(type='YOLOv5KeepRatioResize', scale=(
|
| 446 |
+
320,
|
| 447 |
+
320,
|
| 448 |
+
)),
|
| 449 |
+
dict(
|
| 450 |
+
type='LetterResize',
|
| 451 |
+
scale=(
|
| 452 |
+
320,
|
| 453 |
+
320,
|
| 454 |
+
),
|
| 455 |
+
allow_scale_up=False,
|
| 456 |
+
pad_val=dict(img=114)),
|
| 457 |
+
]),
|
| 458 |
+
dict(
|
| 459 |
+
type='Compose',
|
| 460 |
+
transforms=[
|
| 461 |
+
dict(type='YOLOv5KeepRatioResize', scale=(
|
| 462 |
+
960,
|
| 463 |
+
960,
|
| 464 |
+
)),
|
| 465 |
+
dict(
|
| 466 |
+
type='LetterResize',
|
| 467 |
+
scale=(
|
| 468 |
+
960,
|
| 469 |
+
960,
|
| 470 |
+
),
|
| 471 |
+
allow_scale_up=False,
|
| 472 |
+
pad_val=dict(img=114)),
|
| 473 |
+
]),
|
| 474 |
+
]
|
| 475 |
+
tta_pipeline = [
|
| 476 |
+
dict(type='LoadImageFromFile', file_client_args=dict(backend='disk')),
|
| 477 |
+
dict(
|
| 478 |
+
type='TestTimeAug',
|
| 479 |
+
transforms=[
|
| 480 |
+
[
|
| 481 |
+
dict(
|
| 482 |
+
type='Compose',
|
| 483 |
+
transforms=[
|
| 484 |
+
dict(type='YOLOv5KeepRatioResize', scale=(
|
| 485 |
+
640,
|
| 486 |
+
640,
|
| 487 |
+
)),
|
| 488 |
+
dict(
|
| 489 |
+
type='LetterResize',
|
| 490 |
+
scale=(
|
| 491 |
+
640,
|
| 492 |
+
640,
|
| 493 |
+
),
|
| 494 |
+
allow_scale_up=False,
|
| 495 |
+
pad_val=dict(img=114)),
|
| 496 |
+
]),
|
| 497 |
+
dict(
|
| 498 |
+
type='Compose',
|
| 499 |
+
transforms=[
|
| 500 |
+
dict(type='YOLOv5KeepRatioResize', scale=(
|
| 501 |
+
320,
|
| 502 |
+
320,
|
| 503 |
+
)),
|
| 504 |
+
dict(
|
| 505 |
+
type='LetterResize',
|
| 506 |
+
scale=(
|
| 507 |
+
320,
|
| 508 |
+
320,
|
| 509 |
+
),
|
| 510 |
+
allow_scale_up=False,
|
| 511 |
+
pad_val=dict(img=114)),
|
| 512 |
+
]),
|
| 513 |
+
dict(
|
| 514 |
+
type='Compose',
|
| 515 |
+
transforms=[
|
| 516 |
+
dict(type='YOLOv5KeepRatioResize', scale=(
|
| 517 |
+
960,
|
| 518 |
+
960,
|
| 519 |
+
)),
|
| 520 |
+
dict(
|
| 521 |
+
type='LetterResize',
|
| 522 |
+
scale=(
|
| 523 |
+
960,
|
| 524 |
+
960,
|
| 525 |
+
),
|
| 526 |
+
allow_scale_up=False,
|
| 527 |
+
pad_val=dict(img=114)),
|
| 528 |
+
]),
|
| 529 |
+
],
|
| 530 |
+
[
|
| 531 |
+
dict(type='mmdet.RandomFlip', prob=1.0),
|
| 532 |
+
dict(type='mmdet.RandomFlip', prob=0.0),
|
| 533 |
+
],
|
| 534 |
+
[
|
| 535 |
+
dict(type='mmdet.LoadAnnotations', with_bbox=True),
|
| 536 |
+
],
|
| 537 |
+
[
|
| 538 |
+
dict(
|
| 539 |
+
type='mmdet.PackDetInputs',
|
| 540 |
+
meta_keys=(
|
| 541 |
+
'img_id',
|
| 542 |
+
'img_path',
|
| 543 |
+
'ori_shape',
|
| 544 |
+
'img_shape',
|
| 545 |
+
'scale_factor',
|
| 546 |
+
'pad_param',
|
| 547 |
+
'flip',
|
| 548 |
+
'flip_direction',
|
| 549 |
+
)),
|
| 550 |
+
],
|
| 551 |
+
]),
|
| 552 |
+
]
|
| 553 |
+
|
| 554 |
+
launcher = 'none'
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/custom_dataset/yolov7_l_syncbn_fast_8x16b-300e_coco.py
ADDED
|
@@ -0,0 +1,472 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = ['../_base_/default_runtime.py', '../_base_/det_p5_tta.py']
|
| 2 |
+
|
| 3 |
+
data_root = './data-df2/'
|
| 4 |
+
train_ann_file = 'annotations/instances_train2017.json'
|
| 5 |
+
train_data_prefix = 'train2017/'
|
| 6 |
+
val_ann_file = 'annotations/instances_val2017.json'
|
| 7 |
+
val_data_prefix = 'val2017/'
|
| 8 |
+
num_classes = 13
|
| 9 |
+
train_batch_size_per_gpu = 16
|
| 10 |
+
train_num_workers = 8
|
| 11 |
+
persistent_workers = True
|
| 12 |
+
|
| 13 |
+
vis_backends = [
|
| 14 |
+
dict(type='LocalVisBackend'),
|
| 15 |
+
]
|
| 16 |
+
visualizer = dict(
|
| 17 |
+
type='mmdet.DetLocalVisualizer',
|
| 18 |
+
vis_backends=[
|
| 19 |
+
dict(type='LocalVisBackend'),
|
| 20 |
+
dict(type='WandbVisBackend')
|
| 21 |
+
],
|
| 22 |
+
name='visualizer')
|
| 23 |
+
log_processor = dict(type='LogProcessor', window_size=50, by_epoch=True)
|
| 24 |
+
log_level = 'INFO'
|
| 25 |
+
load_from = None
|
| 26 |
+
resume = False
|
| 27 |
+
|
| 28 |
+
anchors = [
|
| 29 |
+
[(12, 16), (19, 36), (40, 28)], # P3/8
|
| 30 |
+
[(36, 75), (76, 55), (72, 146)], # P4/16
|
| 31 |
+
[(142, 110), (192, 243), (459, 401)] # P5/32
|
| 32 |
+
]
|
| 33 |
+
|
| 34 |
+
base_lr = 0.01
|
| 35 |
+
max_epochs = 100
|
| 36 |
+
|
| 37 |
+
num_epoch_stage2 = 10 # The last 10 epochs switch evaluation interval
|
| 38 |
+
val_interval_stage2 = 1
|
| 39 |
+
|
| 40 |
+
model_test_cfg = dict(
|
| 41 |
+
multi_label=True,
|
| 42 |
+
nms_pre=30000,
|
| 43 |
+
score_thr=0.001,
|
| 44 |
+
nms=dict(type='nms', iou_threshold=0.65),
|
| 45 |
+
max_per_img=300)
|
| 46 |
+
|
| 47 |
+
img_scale = (640, 640)
|
| 48 |
+
dataset_type = 'YOLOv5CocoDataset'
|
| 49 |
+
val_batch_size_per_gpu = 1
|
| 50 |
+
val_num_workers = 2
|
| 51 |
+
batch_shapes_cfg = dict(
|
| 52 |
+
type='BatchShapePolicy',
|
| 53 |
+
batch_size=val_batch_size_per_gpu,
|
| 54 |
+
img_size=img_scale[0],
|
| 55 |
+
size_divisor=32,
|
| 56 |
+
extra_pad_ratio=0.5)
|
| 57 |
+
strides = [8, 16, 32] # Strides of multi-scale prior box
|
| 58 |
+
num_det_layers = 3
|
| 59 |
+
norm_cfg = dict(type='BN', momentum=0.03, eps=0.001)
|
| 60 |
+
|
| 61 |
+
# Data augmentation
|
| 62 |
+
max_translate_ratio = 0.2 # YOLOv5RandomAffine
|
| 63 |
+
scaling_ratio_range = (0.1, 2.0) # YOLOv5RandomAffine
|
| 64 |
+
mixup_prob = 0.15 # YOLOv5MixUp
|
| 65 |
+
randchoice_mosaic_prob = [0.8, 0.2]
|
| 66 |
+
mixup_alpha = 8.0 # YOLOv5MixUp
|
| 67 |
+
mixup_beta = 8.0 # YOLOv5MixUp
|
| 68 |
+
|
| 69 |
+
# -----train val related-----
|
| 70 |
+
loss_cls_weight = 0.3
|
| 71 |
+
loss_bbox_weight = 0.05
|
| 72 |
+
loss_obj_weight = 0.7
|
| 73 |
+
# BatchYOLOv7Assigner params
|
| 74 |
+
simota_candidate_topk = 10
|
| 75 |
+
simota_iou_weight = 3.0
|
| 76 |
+
simota_cls_weight = 1.0
|
| 77 |
+
prior_match_thr = 4. # Priori box matching threshold
|
| 78 |
+
obj_level_weights = [4., 1.,
|
| 79 |
+
0.4] # The obj loss weights of the three output layers
|
| 80 |
+
|
| 81 |
+
lr_factor = 0.1 # Learning rate scaling factor
|
| 82 |
+
weight_decay = 0.0005
|
| 83 |
+
save_epoch_intervals = 2
|
| 84 |
+
max_keep_ckpts = 5
|
| 85 |
+
|
| 86 |
+
env_cfg = dict(
|
| 87 |
+
cudnn_benchmark=True,
|
| 88 |
+
mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0),
|
| 89 |
+
dist_cfg=dict(backend='nccl'))
|
| 90 |
+
|
| 91 |
+
# ===============================Unmodified in most cases====================
|
| 92 |
+
model = dict(
|
| 93 |
+
type='YOLODetector',
|
| 94 |
+
data_preprocessor=dict(
|
| 95 |
+
type='YOLOv5DetDataPreprocessor',
|
| 96 |
+
mean=[0., 0., 0.],
|
| 97 |
+
std=[255., 255., 255.],
|
| 98 |
+
bgr_to_rgb=True),
|
| 99 |
+
backbone=dict(
|
| 100 |
+
type='YOLOv7Backbone',
|
| 101 |
+
arch='L',
|
| 102 |
+
norm_cfg=norm_cfg,
|
| 103 |
+
act_cfg=dict(type='SiLU', inplace=True)),
|
| 104 |
+
neck=dict(
|
| 105 |
+
type='YOLOv7PAFPN',
|
| 106 |
+
block_cfg=dict(
|
| 107 |
+
type='ELANBlock',
|
| 108 |
+
middle_ratio=0.5,
|
| 109 |
+
block_ratio=0.25,
|
| 110 |
+
num_blocks=4,
|
| 111 |
+
num_convs_in_block=1),
|
| 112 |
+
upsample_feats_cat_first=False,
|
| 113 |
+
in_channels=[512, 1024, 1024],
|
| 114 |
+
# The real output channel will be multiplied by 2
|
| 115 |
+
out_channels=[128, 256, 512],
|
| 116 |
+
norm_cfg=norm_cfg,
|
| 117 |
+
act_cfg=dict(type='SiLU', inplace=True)),
|
| 118 |
+
bbox_head=dict(
|
| 119 |
+
type='YOLOv7Head',
|
| 120 |
+
head_module=dict(
|
| 121 |
+
type='YOLOv7HeadModule',
|
| 122 |
+
num_classes=num_classes,
|
| 123 |
+
in_channels=[256, 512, 1024],
|
| 124 |
+
featmap_strides=strides,
|
| 125 |
+
num_base_priors=3),
|
| 126 |
+
prior_generator=dict(
|
| 127 |
+
type='mmdet.YOLOAnchorGenerator',
|
| 128 |
+
base_sizes=anchors,
|
| 129 |
+
strides=strides),
|
| 130 |
+
# scaled based on number of detection layers
|
| 131 |
+
loss_cls=dict(
|
| 132 |
+
type='mmdet.CrossEntropyLoss',
|
| 133 |
+
use_sigmoid=True,
|
| 134 |
+
reduction='mean',
|
| 135 |
+
loss_weight=loss_cls_weight *
|
| 136 |
+
(num_classes / 80 * 3 / num_det_layers)),
|
| 137 |
+
loss_bbox=dict(
|
| 138 |
+
type='IoULoss',
|
| 139 |
+
iou_mode='ciou',
|
| 140 |
+
bbox_format='xywh',
|
| 141 |
+
reduction='mean',
|
| 142 |
+
loss_weight=loss_bbox_weight * (3 / num_det_layers),
|
| 143 |
+
return_iou=True),
|
| 144 |
+
loss_obj=dict(
|
| 145 |
+
type='mmdet.CrossEntropyLoss',
|
| 146 |
+
use_sigmoid=True,
|
| 147 |
+
reduction='mean',
|
| 148 |
+
loss_weight=loss_obj_weight *
|
| 149 |
+
((img_scale[0] / 640)**2 * 3 / num_det_layers)),
|
| 150 |
+
prior_match_thr=prior_match_thr,
|
| 151 |
+
obj_level_weights=obj_level_weights,
|
| 152 |
+
# BatchYOLOv7Assigner params
|
| 153 |
+
simota_candidate_topk=simota_candidate_topk,
|
| 154 |
+
simota_iou_weight=simota_iou_weight,
|
| 155 |
+
simota_cls_weight=simota_cls_weight),
|
| 156 |
+
test_cfg=model_test_cfg)
|
| 157 |
+
|
| 158 |
+
pre_transform = [
|
| 159 |
+
dict(type='LoadImageFromFile', file_client_args=_base_.file_client_args),
|
| 160 |
+
dict(type='LoadAnnotations', with_bbox=True)
|
| 161 |
+
]
|
| 162 |
+
|
| 163 |
+
mosiac4_pipeline = [
|
| 164 |
+
dict(
|
| 165 |
+
type='Mosaic',
|
| 166 |
+
img_scale=img_scale,
|
| 167 |
+
pad_val=114.0,
|
| 168 |
+
pre_transform=pre_transform),
|
| 169 |
+
dict(
|
| 170 |
+
type='YOLOv5RandomAffine',
|
| 171 |
+
max_rotate_degree=0.0,
|
| 172 |
+
max_shear_degree=0.0,
|
| 173 |
+
max_translate_ratio=max_translate_ratio, # note
|
| 174 |
+
scaling_ratio_range=scaling_ratio_range, # note
|
| 175 |
+
# img_scale is (width, height)
|
| 176 |
+
border=(-img_scale[0] // 2, -img_scale[1] // 2),
|
| 177 |
+
border_val=(114, 114, 114)),
|
| 178 |
+
]
|
| 179 |
+
|
| 180 |
+
mosiac9_pipeline = [
|
| 181 |
+
dict(
|
| 182 |
+
type='Mosaic9',
|
| 183 |
+
img_scale=img_scale,
|
| 184 |
+
pad_val=114.0,
|
| 185 |
+
pre_transform=pre_transform),
|
| 186 |
+
dict(
|
| 187 |
+
type='YOLOv5RandomAffine',
|
| 188 |
+
max_rotate_degree=0.0,
|
| 189 |
+
max_shear_degree=0.0,
|
| 190 |
+
max_translate_ratio=max_translate_ratio, # note
|
| 191 |
+
scaling_ratio_range=scaling_ratio_range, # note
|
| 192 |
+
# img_scale is (width, height)
|
| 193 |
+
border=(-img_scale[0] // 2, -img_scale[1] // 2),
|
| 194 |
+
border_val=(114, 114, 114)),
|
| 195 |
+
]
|
| 196 |
+
|
| 197 |
+
randchoice_mosaic_pipeline = dict(
|
| 198 |
+
type='RandomChoice',
|
| 199 |
+
transforms=[mosiac4_pipeline, mosiac9_pipeline],
|
| 200 |
+
prob=randchoice_mosaic_prob)
|
| 201 |
+
|
| 202 |
+
train_pipeline = [
|
| 203 |
+
*pre_transform,
|
| 204 |
+
randchoice_mosaic_pipeline,
|
| 205 |
+
dict(
|
| 206 |
+
type='YOLOv5MixUp',
|
| 207 |
+
alpha=mixup_alpha, # note
|
| 208 |
+
beta=mixup_beta, # note
|
| 209 |
+
prob=mixup_prob,
|
| 210 |
+
pre_transform=[*pre_transform, randchoice_mosaic_pipeline]),
|
| 211 |
+
dict(type='YOLOv5HSVRandomAug'),
|
| 212 |
+
dict(type='mmdet.RandomFlip', prob=0.5),
|
| 213 |
+
dict(
|
| 214 |
+
type='mmdet.PackDetInputs',
|
| 215 |
+
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape', 'flip',
|
| 216 |
+
'flip_direction'))
|
| 217 |
+
]
|
| 218 |
+
|
| 219 |
+
train_dataloader = dict(
|
| 220 |
+
batch_size=train_batch_size_per_gpu,
|
| 221 |
+
num_workers=train_num_workers,
|
| 222 |
+
persistent_workers=persistent_workers,
|
| 223 |
+
pin_memory=True,
|
| 224 |
+
sampler=dict(type='DefaultSampler', shuffle=True),
|
| 225 |
+
collate_fn=dict(type='yolov5_collate'), # FASTER
|
| 226 |
+
dataset=dict(
|
| 227 |
+
type=dataset_type,
|
| 228 |
+
data_root=data_root,
|
| 229 |
+
ann_file=train_ann_file,
|
| 230 |
+
data_prefix=dict(img=train_data_prefix),
|
| 231 |
+
filter_cfg=dict(filter_empty_gt=False, min_size=32),
|
| 232 |
+
pipeline=train_pipeline))
|
| 233 |
+
|
| 234 |
+
test_pipeline = [
|
| 235 |
+
dict(type='LoadImageFromFile', file_client_args=_base_.file_client_args),
|
| 236 |
+
dict(type='YOLOv5KeepRatioResize', scale=img_scale),
|
| 237 |
+
dict(
|
| 238 |
+
type='LetterResize',
|
| 239 |
+
scale=img_scale,
|
| 240 |
+
allow_scale_up=False,
|
| 241 |
+
pad_val=dict(img=114)),
|
| 242 |
+
dict(type='LoadAnnotations', with_bbox=True, _scope_='mmdet'),
|
| 243 |
+
dict(
|
| 244 |
+
type='mmdet.PackDetInputs',
|
| 245 |
+
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
|
| 246 |
+
'scale_factor', 'pad_param'))
|
| 247 |
+
]
|
| 248 |
+
|
| 249 |
+
val_dataloader = dict(
|
| 250 |
+
batch_size=val_batch_size_per_gpu,
|
| 251 |
+
num_workers=val_num_workers,
|
| 252 |
+
persistent_workers=persistent_workers,
|
| 253 |
+
pin_memory=True,
|
| 254 |
+
drop_last=False,
|
| 255 |
+
sampler=dict(type='DefaultSampler', shuffle=False),
|
| 256 |
+
dataset=dict(
|
| 257 |
+
type=dataset_type,
|
| 258 |
+
data_root=data_root,
|
| 259 |
+
test_mode=True,
|
| 260 |
+
data_prefix=dict(img=val_data_prefix),
|
| 261 |
+
ann_file=val_ann_file,
|
| 262 |
+
pipeline=test_pipeline,
|
| 263 |
+
batch_shapes_cfg=batch_shapes_cfg))
|
| 264 |
+
|
| 265 |
+
test_dataloader = val_dataloader
|
| 266 |
+
|
| 267 |
+
param_scheduler = None
|
| 268 |
+
optim_wrapper = dict(
|
| 269 |
+
type='OptimWrapper',
|
| 270 |
+
optimizer=dict(
|
| 271 |
+
type='SGD',
|
| 272 |
+
lr=base_lr,
|
| 273 |
+
momentum=0.937,
|
| 274 |
+
weight_decay=weight_decay,
|
| 275 |
+
nesterov=True,
|
| 276 |
+
batch_size_per_gpu=train_batch_size_per_gpu),
|
| 277 |
+
constructor='YOLOv7OptimWrapperConstructor')
|
| 278 |
+
|
| 279 |
+
default_scope = 'mmyolo'
|
| 280 |
+
default_hooks = dict(
|
| 281 |
+
timer=dict(type='IterTimerHook'),
|
| 282 |
+
logger=dict(type='LoggerHook', interval=2),
|
| 283 |
+
param_scheduler=dict(
|
| 284 |
+
type='YOLOv5ParamSchedulerHook',
|
| 285 |
+
scheduler_type='cosine',
|
| 286 |
+
lr_factor=lr_factor, # note
|
| 287 |
+
max_epochs=max_epochs),
|
| 288 |
+
checkpoint=dict(
|
| 289 |
+
type='CheckpointHook',
|
| 290 |
+
save_param_scheduler=False,
|
| 291 |
+
interval=save_epoch_intervals,
|
| 292 |
+
save_best='auto',
|
| 293 |
+
max_keep_ckpts=max_keep_ckpts),
|
| 294 |
+
sampler_seed=dict(type='DistSamplerSeedHook'),
|
| 295 |
+
visualization=dict(type='mmdet.DetVisualizationHook'))
|
| 296 |
+
|
| 297 |
+
custom_hooks = [
|
| 298 |
+
dict(
|
| 299 |
+
type='EMAHook',
|
| 300 |
+
ema_type='ExpMomentumEMA',
|
| 301 |
+
momentum=0.0001,
|
| 302 |
+
update_buffers=True,
|
| 303 |
+
strict_load=False,
|
| 304 |
+
priority=49)
|
| 305 |
+
]
|
| 306 |
+
|
| 307 |
+
val_evaluator = dict(
|
| 308 |
+
type='mmdet.CocoMetric',
|
| 309 |
+
proposal_nums=(100, 1, 10), # Can be accelerated
|
| 310 |
+
ann_file=data_root + val_ann_file,
|
| 311 |
+
metric='bbox')
|
| 312 |
+
test_evaluator = val_evaluator
|
| 313 |
+
|
| 314 |
+
train_cfg = dict(
|
| 315 |
+
type='EpochBasedTrainLoop',
|
| 316 |
+
max_epochs=max_epochs,
|
| 317 |
+
val_interval=save_epoch_intervals,
|
| 318 |
+
dynamic_intervals=[(max_epochs - num_epoch_stage2, val_interval_stage2)])
|
| 319 |
+
val_cfg = dict(type='ValLoop')
|
| 320 |
+
test_cfg = dict(type='TestLoop')
|
| 321 |
+
|
| 322 |
+
# ============================
|
| 323 |
+
|
| 324 |
+
file_client_args = dict(backend='disk')
|
| 325 |
+
_file_client_args = dict(backend='disk')
|
| 326 |
+
tta_model = dict(
|
| 327 |
+
type='mmdet.DetTTAModel',
|
| 328 |
+
tta_cfg=dict(nms=dict(type='nms', iou_threshold=0.65), max_per_img=300))
|
| 329 |
+
img_scales = [
|
| 330 |
+
(
|
| 331 |
+
640,
|
| 332 |
+
640,
|
| 333 |
+
),
|
| 334 |
+
(
|
| 335 |
+
320,
|
| 336 |
+
320,
|
| 337 |
+
),
|
| 338 |
+
(
|
| 339 |
+
960,
|
| 340 |
+
960,
|
| 341 |
+
),
|
| 342 |
+
]
|
| 343 |
+
_multiscale_resize_transforms = [
|
| 344 |
+
dict(
|
| 345 |
+
type='Compose',
|
| 346 |
+
transforms=[
|
| 347 |
+
dict(type='YOLOv5KeepRatioResize', scale=(
|
| 348 |
+
640,
|
| 349 |
+
640,
|
| 350 |
+
)),
|
| 351 |
+
dict(
|
| 352 |
+
type='LetterResize',
|
| 353 |
+
scale=(
|
| 354 |
+
640,
|
| 355 |
+
640,
|
| 356 |
+
),
|
| 357 |
+
allow_scale_up=False,
|
| 358 |
+
pad_val=dict(img=114)),
|
| 359 |
+
]),
|
| 360 |
+
dict(
|
| 361 |
+
type='Compose',
|
| 362 |
+
transforms=[
|
| 363 |
+
dict(type='YOLOv5KeepRatioResize', scale=(
|
| 364 |
+
320,
|
| 365 |
+
320,
|
| 366 |
+
)),
|
| 367 |
+
dict(
|
| 368 |
+
type='LetterResize',
|
| 369 |
+
scale=(
|
| 370 |
+
320,
|
| 371 |
+
320,
|
| 372 |
+
),
|
| 373 |
+
allow_scale_up=False,
|
| 374 |
+
pad_val=dict(img=114)),
|
| 375 |
+
]),
|
| 376 |
+
dict(
|
| 377 |
+
type='Compose',
|
| 378 |
+
transforms=[
|
| 379 |
+
dict(type='YOLOv5KeepRatioResize', scale=(
|
| 380 |
+
960,
|
| 381 |
+
960,
|
| 382 |
+
)),
|
| 383 |
+
dict(
|
| 384 |
+
type='LetterResize',
|
| 385 |
+
scale=(
|
| 386 |
+
960,
|
| 387 |
+
960,
|
| 388 |
+
),
|
| 389 |
+
allow_scale_up=False,
|
| 390 |
+
pad_val=dict(img=114)),
|
| 391 |
+
]),
|
| 392 |
+
]
|
| 393 |
+
tta_pipeline = [
|
| 394 |
+
dict(type='LoadImageFromFile', file_client_args=dict(backend='disk')),
|
| 395 |
+
dict(
|
| 396 |
+
type='TestTimeAug',
|
| 397 |
+
transforms=[
|
| 398 |
+
[
|
| 399 |
+
dict(
|
| 400 |
+
type='Compose',
|
| 401 |
+
transforms=[
|
| 402 |
+
dict(type='YOLOv5KeepRatioResize', scale=(
|
| 403 |
+
640,
|
| 404 |
+
640,
|
| 405 |
+
)),
|
| 406 |
+
dict(
|
| 407 |
+
type='LetterResize',
|
| 408 |
+
scale=(
|
| 409 |
+
640,
|
| 410 |
+
640,
|
| 411 |
+
),
|
| 412 |
+
allow_scale_up=False,
|
| 413 |
+
pad_val=dict(img=114)),
|
| 414 |
+
]),
|
| 415 |
+
dict(
|
| 416 |
+
type='Compose',
|
| 417 |
+
transforms=[
|
| 418 |
+
dict(type='YOLOv5KeepRatioResize', scale=(
|
| 419 |
+
320,
|
| 420 |
+
320,
|
| 421 |
+
)),
|
| 422 |
+
dict(
|
| 423 |
+
type='LetterResize',
|
| 424 |
+
scale=(
|
| 425 |
+
320,
|
| 426 |
+
320,
|
| 427 |
+
),
|
| 428 |
+
allow_scale_up=False,
|
| 429 |
+
pad_val=dict(img=114)),
|
| 430 |
+
]),
|
| 431 |
+
dict(
|
| 432 |
+
type='Compose',
|
| 433 |
+
transforms=[
|
| 434 |
+
dict(type='YOLOv5KeepRatioResize', scale=(
|
| 435 |
+
960,
|
| 436 |
+
960,
|
| 437 |
+
)),
|
| 438 |
+
dict(
|
| 439 |
+
type='LetterResize',
|
| 440 |
+
scale=(
|
| 441 |
+
960,
|
| 442 |
+
960,
|
| 443 |
+
),
|
| 444 |
+
allow_scale_up=False,
|
| 445 |
+
pad_val=dict(img=114)),
|
| 446 |
+
]),
|
| 447 |
+
],
|
| 448 |
+
[
|
| 449 |
+
dict(type='mmdet.RandomFlip', prob=1.0),
|
| 450 |
+
dict(type='mmdet.RandomFlip', prob=0.0),
|
| 451 |
+
],
|
| 452 |
+
[
|
| 453 |
+
dict(type='mmdet.LoadAnnotations', with_bbox=True),
|
| 454 |
+
],
|
| 455 |
+
[
|
| 456 |
+
dict(
|
| 457 |
+
type='mmdet.PackDetInputs',
|
| 458 |
+
meta_keys=(
|
| 459 |
+
'img_id',
|
| 460 |
+
'img_path',
|
| 461 |
+
'ori_shape',
|
| 462 |
+
'img_shape',
|
| 463 |
+
'scale_factor',
|
| 464 |
+
'pad_param',
|
| 465 |
+
'flip',
|
| 466 |
+
'flip_direction',
|
| 467 |
+
)),
|
| 468 |
+
],
|
| 469 |
+
]),
|
| 470 |
+
]
|
| 471 |
+
|
| 472 |
+
launcher = 'none'
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/README.md
ADDED
|
@@ -0,0 +1,118 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# YOLOv5
|
| 2 |
+
|
| 3 |
+
<!-- [ALGORITHM] -->
|
| 4 |
+
|
| 5 |
+
## Abstract
|
| 6 |
+
|
| 7 |
+
YOLOv5 is a family of object detection architectures and models pretrained on the COCO dataset, and represents Ultralytics open-source research into future vision AI methods, incorporating lessons learned and best practices evolved over thousands of hours of research and development.
|
| 8 |
+
|
| 9 |
+
<div align=center>
|
| 10 |
+
<img src="https://user-images.githubusercontent.com/27466624/200000324-70ae078f-cea7-4189-8baa-440656797dad.jpg"/>
|
| 11 |
+
YOLOv5-l-P5 model structure
|
| 12 |
+
</div>
|
| 13 |
+
|
| 14 |
+
<div align=center>
|
| 15 |
+
<img src="https://user-images.githubusercontent.com/27466624/211143533-1725c1b2-6189-4c3a-a046-ad968e03cb9d.jpg"/>
|
| 16 |
+
YOLOv5-l-P6 model structure
|
| 17 |
+
</div>
|
| 18 |
+
|
| 19 |
+
## Results and models
|
| 20 |
+
|
| 21 |
+
### COCO
|
| 22 |
+
|
| 23 |
+
| Backbone | Arch | size | SyncBN | AMP | Mem (GB) | box AP | TTA box AP | Config | Download |
|
| 24 |
+
| :------: | :--: | :--: | :----: | :-: | :------: | :----: | :--------: | :--------------------------------------------------------------------------------------------------------------------: | :------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
|
| 25 |
+
| YOLOv5-n | P5 | 640 | Yes | Yes | 1.5 | 28.0 | 30.7 | [config](https://github.com/open-mmlab/mmyolo/tree/main/configs/yolov5/yolov5_n-v61_syncbn_fast_8xb16-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_n-v61_syncbn_fast_8xb16-300e_coco/yolov5_n-v61_syncbn_fast_8xb16-300e_coco_20220919_090739-b804c1ad.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_n-v61_syncbn_fast_8xb16-300e_coco/yolov5_n-v61_syncbn_fast_8xb16-300e_coco_20220919_090739.log.json) |
|
| 26 |
+
| YOLOv5-s | P5 | 640 | Yes | Yes | 2.7 | 37.7 | 40.2 | [config](https://github.com/open-mmlab/mmyolo/tree/main/configs/yolov5/yolov5_s-v61_syncbn_fast_8xb16-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_s-v61_syncbn_fast_8xb16-300e_coco/yolov5_s-v61_syncbn_fast_8xb16-300e_coco_20220918_084700-86e02187.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_s-v61_syncbn_fast_8xb16-300e_coco/yolov5_s-v61_syncbn_fast_8xb16-300e_coco_20220918_084700.log.json) |
|
| 27 |
+
| YOLOv5-m | P5 | 640 | Yes | Yes | 5.0 | 45.3 | 46.9 | [config](https://github.com/open-mmlab/mmyolo/tree/main/configs/yolov5/yolov5_m-v61_syncbn_fast_8xb16-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_m-v61_syncbn_fast_8xb16-300e_coco/yolov5_m-v61_syncbn_fast_8xb16-300e_coco_20220917_204944-516a710f.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_m-v61_syncbn_fast_8xb16-300e_coco/yolov5_m-v61_syncbn_fast_8xb16-300e_coco_20220917_204944.log.json) |
|
| 28 |
+
| YOLOv5-l | P5 | 640 | Yes | Yes | 8.1 | 48.8 | 49.9 | [config](https://github.com/open-mmlab/mmyolo/tree/main/configs/yolov5/yolov5_l-v61_syncbn_fast_8xb16-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_l-v61_syncbn_fast_8xb16-300e_coco/yolov5_l-v61_syncbn_fast_8xb16-300e_coco_20220917_031007-096ef0eb.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_l-v61_syncbn_fast_8xb16-300e_coco/yolov5_l-v61_syncbn_fast_8xb16-300e_coco_20220917_031007.log.json) |
|
| 29 |
+
| YOLOv5-n | P6 | 1280 | Yes | Yes | 5.8 | 35.9 | | [config](https://github.com/open-mmlab/mmyolo/tree/main/configs/yolov5/yolov5_n-p6-v62_syncbn_fast_8xb16-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_n-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_n-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_224705-d493c5f3.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_n-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_n-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_224705.log.json) |
|
| 30 |
+
| YOLOv5-s | P6 | 1280 | Yes | Yes | 10.5 | 44.4 | | [config](https://github.com/open-mmlab/mmyolo/tree/main/configs/yolov5/yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_215044-58865c19.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_215044.log.json) |
|
| 31 |
+
| YOLOv5-m | P6 | 1280 | Yes | Yes | 19.1 | 51.3 | | [config](https://github.com/open-mmlab/mmyolo/tree/main/configs/yolov5/yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_230453-49564d58.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_230453.log.json) |
|
| 32 |
+
| YOLOv5-l | P6 | 1280 | Yes | Yes | 30.5 | 53.7 | | [config](https://github.com/open-mmlab/mmyolo/tree/main/configs/yolov5/yolov5_l-p6-v62_syncbn_fast_8xb16-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_l-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_l-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_234308-7a2ba6bf.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_l-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_l-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_234308.log.json) |
|
| 33 |
+
|
| 34 |
+
**Note**:
|
| 35 |
+
In the official YOLOv5 code, the `random_perspective` data augmentation in COCO object detection task training uses mask annotation information, which leads to higher performance. Object detection should not use mask annotation, so only box annotation information is used in `MMYOLO`. We will use the mask annotation information in the instance segmentation task. See https://github.com/ultralytics/yolov5/issues/9917 for details.
|
| 36 |
+
|
| 37 |
+
1. `fast` means that `YOLOv5DetDataPreprocessor` and `yolov5_collate` are used for data preprocessing, which is faster for training, but less flexible for multitasking. Recommended to use fast version config if you only care about object detection.
|
| 38 |
+
2. `detect` means that the network input is fixed to `640x640` and the post-processing thresholds is modified.
|
| 39 |
+
3. `SyncBN` means use SyncBN, `AMP` indicates training with mixed precision.
|
| 40 |
+
4. We use 8x A100 for training, and the single-GPU batch size is 16. This is different from the official code.
|
| 41 |
+
5. The performance is unstable and may fluctuate by about 0.4 mAP and the highest performance weight in `COCO` training in `YOLOv5` may not be the last epoch.
|
| 42 |
+
6. `TTA` means that Test Time Augmentation. It's perform 3 multi-scaling transformations on the image, followed by 2 flipping transformations (flipping and not flipping). You only need to specify `--tta` when testing to enable. see [TTA](https://github.com/open-mmlab/mmyolo/blob/dev/docs/en/common_usage/tta.md) for details.
|
| 43 |
+
|
| 44 |
+
### VOC
|
| 45 |
+
|
| 46 |
+
| Backbone | size | Batchsize | AMP | Mem (GB) | box AP(COCO metric) | Config | Download |
|
| 47 |
+
| :------: | :--: | :-------: | :-: | :------: | :-----------------: | :------------------------------------------------------------------------------------------------------------: | :------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
|
| 48 |
+
| YOLOv5-n | 512 | 64 | Yes | 3.5 | 51.2 | [config](https://github.com/open-mmlab/mmyolo/tree/main/configs/yolov5/voc/yolov5_n-v61_fast_1xb64-50e_voc.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_n-v61_fast_1xb64-50e_voc/yolov5_n-v61_fast_1xb64-50e_voc_20221017_234254-f1493430.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_n-v61_fast_1xb64-50e_voc/yolov5_n-v61_fast_1xb64-50e_voc_20221017_234254.log.json) |
|
| 49 |
+
| YOLOv5-s | 512 | 64 | Yes | 6.5 | 62.7 | [config](https://github.com/open-mmlab/mmyolo/tree/main/configs/yolov5/voc/yolov5_s-v61_fast_1xb64-50e_voc.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_s-v61_fast_1xb64-50e_voc/yolov5_s-v61_fast_1xb64-50e_voc_20221017_234156-0009b33e.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_s-v61_fast_1xb64-50e_voc/yolov5_s-v61_fast_1xb64-50e_voc_20221017_234156.log.json) |
|
| 50 |
+
| YOLOv5-m | 512 | 64 | Yes | 12.0 | 70.1 | [config](https://github.com/open-mmlab/mmyolo/tree/main/configs/yolov5/voc/yolov5_m-v61_fast_1xb64-50e_voc.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_m-v61_fast_1xb64-50e_voc/yolov5_m-v61_fast_1xb64-50e_voc_20221017_114138-815c143a.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_m-v61_fast_1xb64-50e_voc/yolov5_m-v61_fast_1xb64-50e_voc_20221017_114138.log.json) |
|
| 51 |
+
| YOLOv5-l | 512 | 32 | Yes | 10.0 | 73.1 | [config](https://github.com/open-mmlab/mmyolo/tree/main/configs/yolov5/voc/yolov5_l-v61_fast_1xb32-50e_voc.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_l-v61_fast_1xb32-50e_voc/yolov5_l-v61_fast_1xb32-50e_voc_20221017_045500-edc7e0d8.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_l-v61_fast_1xb32-50e_voc/yolov5_l-v61_fast_1xb32-50e_voc_20221017_045500.log.json) |
|
| 52 |
+
|
| 53 |
+
**Note**:
|
| 54 |
+
|
| 55 |
+
1. Training on VOC dataset need pretrained model which trained on COCO.
|
| 56 |
+
2. The performance is unstable and may fluctuate by about 0.4 mAP.
|
| 57 |
+
3. Official YOLOv5 use COCO metric, while training VOC dataset.
|
| 58 |
+
4. We converted the VOC test dataset to COCO format offline, while reproducing mAP result as shown above. We will support to use COCO metric while training VOC dataset in later version.
|
| 59 |
+
5. Hyperparameter reference from `https://wandb.ai/glenn-jocher/YOLOv5_VOC_official`.
|
| 60 |
+
|
| 61 |
+
### CrowdHuman
|
| 62 |
+
|
| 63 |
+
Since the `iscrowd` annotation of the COCO dataset is not equivalent to `ignore`, we use the CrowdHuman dataset to verify that the YOLOv5 ignore logic is correct.
|
| 64 |
+
|
| 65 |
+
| Backbone | size | SyncBN | AMP | Mem (GB) | ignore_iof_thr | box AP50(CrowDHuman Metric) | MR | JI | Config | Download |
|
| 66 |
+
| :------: | :--: | :----: | :-: | :------: | :------------: | :-------------------------: | :--: | :---: | :-----------------------------------------------------------------------------------------------------------------------------: | :------: |
|
| 67 |
+
| YOLOv5-s | 640 | Yes | Yes | 2.6 | -1 | 85.79 | 48.7 | 75.33 | [config](https://github.com/open-mmlab/mmyolo/tree/main/configs/yolov5/crowdhuman/yolov5_s-v61_fast_8xb16-300e_crowdhuman.py) | |
|
| 68 |
+
| YOLOv5-s | 640 | Yes | Yes | 2.6 | 0.5 | 86.17 | 48.8 | 75.87 | [config](https://github.com/open-mmlab/mmyolo/tree/main/configs/yolov5/crowdhuman/yolov5_s-v61_8xb16-300e_ignore_crowdhuman.py) | |
|
| 69 |
+
|
| 70 |
+
**Note**:
|
| 71 |
+
|
| 72 |
+
1. `ignore_iof_thr` is -1 indicating that the ignore tag is not considered. We adjusted with `ignore_iof_thr` thresholds of 0.5, 0.8, 0.9, and the results show that 0.5 has the best performance.
|
| 73 |
+
2. The above table shows the performance of the model with the best performance on the validation set. The best performing models are around 160+ epoch which means that there is no need to train so many epochs.
|
| 74 |
+
3. This is a very simple implementation that simply replaces COCO's anchor with the `tools/analysis_tools/optimize_anchors.py` script. We'll adjust other parameters later to improve performance.
|
| 75 |
+
|
| 76 |
+
## Citation
|
| 77 |
+
|
| 78 |
+
```latex
|
| 79 |
+
@software{glenn_jocher_2022_7002879,
|
| 80 |
+
author = {Glenn Jocher and
|
| 81 |
+
Ayush Chaurasia and
|
| 82 |
+
Alex Stoken and
|
| 83 |
+
Jirka Borovec and
|
| 84 |
+
NanoCode012 and
|
| 85 |
+
Yonghye Kwon and
|
| 86 |
+
TaoXie and
|
| 87 |
+
Kalen Michael and
|
| 88 |
+
Jiacong Fang and
|
| 89 |
+
imyhxy and
|
| 90 |
+
Lorna and
|
| 91 |
+
Colin Wong and
|
| 92 |
+
曾逸夫(Zeng Yifu) and
|
| 93 |
+
Abhiram V and
|
| 94 |
+
Diego Montes and
|
| 95 |
+
Zhiqiang Wang and
|
| 96 |
+
Cristi Fati and
|
| 97 |
+
Jebastin Nadar and
|
| 98 |
+
Laughing and
|
| 99 |
+
UnglvKitDe and
|
| 100 |
+
tkianai and
|
| 101 |
+
yxNONG and
|
| 102 |
+
Piotr Skalski and
|
| 103 |
+
Adam Hogan and
|
| 104 |
+
Max Strobel and
|
| 105 |
+
Mrinal Jain and
|
| 106 |
+
Lorenzo Mammana and
|
| 107 |
+
xylieong},
|
| 108 |
+
title = {{ultralytics/yolov5: v6.2 - YOLOv5 Classification
|
| 109 |
+
Models, Apple M1, Reproducibility, ClearML and
|
| 110 |
+
Deci.ai integrations}},
|
| 111 |
+
month = aug,
|
| 112 |
+
year = 2022,
|
| 113 |
+
publisher = {Zenodo},
|
| 114 |
+
version = {v6.2},
|
| 115 |
+
doi = {10.5281/zenodo.7002879},
|
| 116 |
+
url = {https://doi.org/10.5281/zenodo.7002879}
|
| 117 |
+
}
|
| 118 |
+
```
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/__init__.py
ADDED
|
File without changes
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/crowdhuman/__init__.py
ADDED
|
File without changes
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/crowdhuman/yolov5_s-v61_8xb16-300e_ignore_crowdhuman.py
ADDED
|
@@ -0,0 +1,63 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = 'yolov5_s-v61_fast_8xb16-300e_crowdhuman.py'
|
| 2 |
+
|
| 3 |
+
model = dict(
|
| 4 |
+
data_preprocessor=dict(
|
| 5 |
+
_delete_=True,
|
| 6 |
+
type='mmdet.DetDataPreprocessor',
|
| 7 |
+
mean=[0., 0., 0.],
|
| 8 |
+
std=[255., 255., 255.],
|
| 9 |
+
bgr_to_rgb=True),
|
| 10 |
+
bbox_head=dict(ignore_iof_thr=0.5))
|
| 11 |
+
|
| 12 |
+
img_scale = _base_.img_scale
|
| 13 |
+
|
| 14 |
+
albu_train_transforms = [
|
| 15 |
+
dict(type='Blur', p=0.01),
|
| 16 |
+
dict(type='MedianBlur', p=0.01),
|
| 17 |
+
dict(type='ToGray', p=0.01),
|
| 18 |
+
dict(type='CLAHE', p=0.01)
|
| 19 |
+
]
|
| 20 |
+
|
| 21 |
+
pre_transform = [
|
| 22 |
+
dict(type='LoadImageFromFile', file_client_args=_base_.file_client_args),
|
| 23 |
+
# only change this
|
| 24 |
+
dict(type='mmdet.LoadAnnotations', with_bbox=True)
|
| 25 |
+
]
|
| 26 |
+
|
| 27 |
+
train_pipeline = [
|
| 28 |
+
*pre_transform,
|
| 29 |
+
dict(
|
| 30 |
+
type='Mosaic',
|
| 31 |
+
img_scale=img_scale,
|
| 32 |
+
pad_val=114.0,
|
| 33 |
+
pre_transform=pre_transform),
|
| 34 |
+
dict(
|
| 35 |
+
type='YOLOv5RandomAffine',
|
| 36 |
+
max_rotate_degree=0.0,
|
| 37 |
+
max_shear_degree=0.0,
|
| 38 |
+
scaling_ratio_range=(0.5, 1.5),
|
| 39 |
+
# img_scale is (width, height)
|
| 40 |
+
border=(-img_scale[0] // 2, -img_scale[1] // 2),
|
| 41 |
+
border_val=(114, 114, 114)),
|
| 42 |
+
dict(
|
| 43 |
+
type='mmdet.Albu',
|
| 44 |
+
transforms=albu_train_transforms,
|
| 45 |
+
bbox_params=dict(
|
| 46 |
+
type='BboxParams',
|
| 47 |
+
format='pascal_voc',
|
| 48 |
+
label_fields=['gt_bboxes_labels', 'gt_ignore_flags']),
|
| 49 |
+
keymap={
|
| 50 |
+
'img': 'image',
|
| 51 |
+
'gt_bboxes': 'bboxes'
|
| 52 |
+
}),
|
| 53 |
+
dict(type='YOLOv5HSVRandomAug'),
|
| 54 |
+
dict(type='mmdet.RandomFlip', prob=0.5),
|
| 55 |
+
dict(
|
| 56 |
+
type='mmdet.PackDetInputs',
|
| 57 |
+
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape', 'flip',
|
| 58 |
+
'flip_direction'))
|
| 59 |
+
]
|
| 60 |
+
|
| 61 |
+
train_dataloader = dict(
|
| 62 |
+
collate_fn=dict(type='pseudo_collate'),
|
| 63 |
+
dataset=dict(pipeline=train_pipeline))
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/crowdhuman/yolov5_s-v61_fast_8xb16-300e_crowdhuman.py
ADDED
|
@@ -0,0 +1,47 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = '../yolov5_s-v61_syncbn_fast_8xb16-300e_coco.py'
|
| 2 |
+
|
| 3 |
+
# Use the model trained on the COCO as the pretrained model
|
| 4 |
+
load_from = 'https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_s-v61_syncbn_fast_8xb16-300e_coco/yolov5_s-v61_syncbn_fast_8xb16-300e_coco_20220918_084700-86e02187.pth' # noqa
|
| 5 |
+
|
| 6 |
+
# dataset settings
|
| 7 |
+
data_root = 'data/CrowdHuman/'
|
| 8 |
+
dataset_type = 'YOLOv5CrowdHumanDataset'
|
| 9 |
+
|
| 10 |
+
# parameters that often need to be modified
|
| 11 |
+
num_classes = 1
|
| 12 |
+
|
| 13 |
+
anchors = [
|
| 14 |
+
[(6, 14), (12, 28), (19, 48)], # P3/8
|
| 15 |
+
[(29, 79), (46, 124), (142, 54)], # P4/16
|
| 16 |
+
[(73, 198), (124, 330), (255, 504)] # P5/32
|
| 17 |
+
]
|
| 18 |
+
|
| 19 |
+
model = dict(
|
| 20 |
+
bbox_head=dict(
|
| 21 |
+
head_module=dict(num_classes=num_classes),
|
| 22 |
+
prior_generator=dict(base_sizes=anchors)))
|
| 23 |
+
|
| 24 |
+
train_dataloader = dict(
|
| 25 |
+
dataset=dict(
|
| 26 |
+
type=dataset_type,
|
| 27 |
+
data_root=data_root,
|
| 28 |
+
ann_file='annotation_train.odgt',
|
| 29 |
+
data_prefix=dict(img='Images/')))
|
| 30 |
+
|
| 31 |
+
val_dataloader = dict(
|
| 32 |
+
dataset=dict(
|
| 33 |
+
type=dataset_type,
|
| 34 |
+
data_root=data_root,
|
| 35 |
+
ann_file='annotation_val.odgt',
|
| 36 |
+
data_prefix=dict(img='Images/'),
|
| 37 |
+
# CrowdHumanMetric does not support out-of-order output images
|
| 38 |
+
# for the time being. batch_shapes_cfg does not support.
|
| 39 |
+
batch_shapes_cfg=None))
|
| 40 |
+
test_dataloader = val_dataloader
|
| 41 |
+
|
| 42 |
+
val_evaluator = dict(
|
| 43 |
+
_delete_=True,
|
| 44 |
+
type='mmdet.CrowdHumanMetric',
|
| 45 |
+
ann_file=data_root + 'annotation_val.odgt',
|
| 46 |
+
metric=['AP', 'MR', 'JI'])
|
| 47 |
+
test_evaluator = val_evaluator
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/metafile.yml
ADDED
|
@@ -0,0 +1,178 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Collections:
|
| 2 |
+
- Name: YOLOv5
|
| 3 |
+
Metadata:
|
| 4 |
+
Training Data: COCO
|
| 5 |
+
Training Techniques:
|
| 6 |
+
- SGD with Nesterov
|
| 7 |
+
- Weight Decay
|
| 8 |
+
- AMP
|
| 9 |
+
- Synchronize BN
|
| 10 |
+
Training Resources: 8x A100 GPUs
|
| 11 |
+
Architecture:
|
| 12 |
+
- CSPDarkNet
|
| 13 |
+
- PAFPN
|
| 14 |
+
README: configs/yolov5/README.md
|
| 15 |
+
Code:
|
| 16 |
+
URL: https://github.com/open-mmlab/mmyolo/blob/v0.1.0/mmyolo/models/detectors/yolo_detector.py#L12
|
| 17 |
+
Version: v0.1.0
|
| 18 |
+
- Name: YOLOv5_VOC
|
| 19 |
+
Metadata:
|
| 20 |
+
Training Data: VOC
|
| 21 |
+
Training Techniques:
|
| 22 |
+
- SGD with Nesterov
|
| 23 |
+
- Weight Decay
|
| 24 |
+
- AMP
|
| 25 |
+
Training Resources: 1x A100 GPU
|
| 26 |
+
Architecture:
|
| 27 |
+
- CSPDarkNet
|
| 28 |
+
- PAFPN
|
| 29 |
+
README: configs/yolov5/README.md
|
| 30 |
+
Code:
|
| 31 |
+
URL: https://github.com/open-mmlab/mmyolo/blob/v0.1.0/mmyolo/models/detectors/yolo_detector.py#L12
|
| 32 |
+
Version: v0.1.0
|
| 33 |
+
|
| 34 |
+
Models:
|
| 35 |
+
- Name: yolov5_n-v61_syncbn_fast_8xb16-300e_coco
|
| 36 |
+
In Collection: YOLOv5
|
| 37 |
+
Config: configs/yolov5/yolov5_n-v61_syncbn_fast_8xb16-300e_coco.py
|
| 38 |
+
Metadata:
|
| 39 |
+
Training Memory (GB): 1.5
|
| 40 |
+
Epochs: 300
|
| 41 |
+
Results:
|
| 42 |
+
- Task: Object Detection
|
| 43 |
+
Dataset: COCO
|
| 44 |
+
Metrics:
|
| 45 |
+
box AP: 28.0
|
| 46 |
+
Weights: https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_n-v61_syncbn_fast_8xb16-300e_coco/yolov5_n-v61_syncbn_fast_8xb16-300e_coco_20220919_090739-b804c1ad.pth
|
| 47 |
+
- Name: yolov5_s-v61_syncbn_fast_8xb16-300e_coco
|
| 48 |
+
In Collection: YOLOv5
|
| 49 |
+
Config: configs/yolov5/yolov5_s-v61_syncbn_fast_8xb16-300e_coco.py
|
| 50 |
+
Metadata:
|
| 51 |
+
Training Memory (GB): 2.7
|
| 52 |
+
Epochs: 300
|
| 53 |
+
Results:
|
| 54 |
+
- Task: Object Detection
|
| 55 |
+
Dataset: COCO
|
| 56 |
+
Metrics:
|
| 57 |
+
box AP: 37.7
|
| 58 |
+
Weights: https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_s-v61_syncbn_fast_8xb16-300e_coco/yolov5_s-v61_syncbn_fast_8xb16-300e_coco_20220918_084700-86e02187.pth
|
| 59 |
+
- Name: yolov5_m-v61_syncbn_fast_8xb16-300e_coco
|
| 60 |
+
In Collection: YOLOv5
|
| 61 |
+
Config: configs/yolov5/yolov5_m-v61_syncbn_fast_8xb16-300e_coco.py
|
| 62 |
+
Metadata:
|
| 63 |
+
Training Memory (GB): 5.0
|
| 64 |
+
Epochs: 300
|
| 65 |
+
Results:
|
| 66 |
+
- Task: Object Detection
|
| 67 |
+
Dataset: COCO
|
| 68 |
+
Metrics:
|
| 69 |
+
box AP: 45.3
|
| 70 |
+
Weights: https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_m-v61_syncbn_fast_8xb16-300e_coco/yolov5_m-v61_syncbn_fast_8xb16-300e_coco_20220917_204944-516a710f.pth
|
| 71 |
+
- Name: yolov5_l-v61_syncbn_fast_8xb16-300e_coco
|
| 72 |
+
In Collection: YOLOv5
|
| 73 |
+
Config: configs/yolov5/yolov5_l-v61_syncbn_fast_8xb16-300e_coco.py
|
| 74 |
+
Metadata:
|
| 75 |
+
Training Memory (GB): 8.1
|
| 76 |
+
Epochs: 300
|
| 77 |
+
Results:
|
| 78 |
+
- Task: Object Detection
|
| 79 |
+
Dataset: COCO
|
| 80 |
+
Metrics:
|
| 81 |
+
box AP: 48.8
|
| 82 |
+
Weights: https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_l-v61_syncbn_fast_8xb16-300e_coco/yolov5_l-v61_syncbn_fast_8xb16-300e_coco_20220917_031007-096ef0eb.pth
|
| 83 |
+
- Name: yolov5_n-p6-v62_syncbn_fast_8xb16-300e_coco
|
| 84 |
+
In Collection: YOLOv5
|
| 85 |
+
Config: configs/yolov5/yolov5_n-p6-v62_syncbn_fast_8xb16-300e_coco.py
|
| 86 |
+
Metadata:
|
| 87 |
+
Training Memory (GB): 5.8
|
| 88 |
+
Epochs: 300
|
| 89 |
+
Results:
|
| 90 |
+
- Task: Object Detection
|
| 91 |
+
Dataset: COCO
|
| 92 |
+
Metrics:
|
| 93 |
+
box AP: 35.9
|
| 94 |
+
Weights: https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_n-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_n-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_224705-d493c5f3.pth
|
| 95 |
+
- Name: yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco
|
| 96 |
+
In Collection: YOLOv5
|
| 97 |
+
Config: configs/yolov5/yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco.py
|
| 98 |
+
Metadata:
|
| 99 |
+
Training Memory (GB): 10.5
|
| 100 |
+
Epochs: 300
|
| 101 |
+
Results:
|
| 102 |
+
- Task: Object Detection
|
| 103 |
+
Dataset: COCO
|
| 104 |
+
Metrics:
|
| 105 |
+
box AP: 44.4
|
| 106 |
+
Weights: https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_215044-58865c19.pth
|
| 107 |
+
- Name: yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco
|
| 108 |
+
In Collection: YOLOv5
|
| 109 |
+
Config: configs/yolov5/yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco.py
|
| 110 |
+
Metadata:
|
| 111 |
+
Training Memory (GB): 19.1
|
| 112 |
+
Epochs: 300
|
| 113 |
+
Results:
|
| 114 |
+
- Task: Object Detection
|
| 115 |
+
Dataset: COCO
|
| 116 |
+
Metrics:
|
| 117 |
+
box AP: 51.3
|
| 118 |
+
Weights: https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_230453-49564d58.pth
|
| 119 |
+
- Name: yolov5_l-p6-v62_syncbn_fast_8xb16-300e_coco
|
| 120 |
+
In Collection: YOLOv5
|
| 121 |
+
Config: configs/yolov5/yolov5_l-p6-v62_syncbn_fast_8xb16-300e_coco.py
|
| 122 |
+
Metadata:
|
| 123 |
+
Training Memory (GB): 30.5
|
| 124 |
+
Epochs: 300
|
| 125 |
+
Results:
|
| 126 |
+
- Task: Object Detection
|
| 127 |
+
Dataset: COCO
|
| 128 |
+
Metrics:
|
| 129 |
+
box AP: 53.7
|
| 130 |
+
Weights: https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_l-p6-v62_syncbn_fast_8xb16-300e_coco/yolov5_l-p6-v62_syncbn_fast_8xb16-300e_coco_20221027_234308-7a2ba6bf.pth
|
| 131 |
+
- Name: yolov5_n-v61_fast_1xb64-50e_voc
|
| 132 |
+
In Collection: YOLOv5_VOC
|
| 133 |
+
Config: configs/yolov5/voc/yolov5_n-v61_fast_1xb64-50e_voc.py
|
| 134 |
+
Metadata:
|
| 135 |
+
Training Memory (GB): 3.5
|
| 136 |
+
Epochs: 50
|
| 137 |
+
Results:
|
| 138 |
+
- Task: Object Detection
|
| 139 |
+
Dataset: COCO
|
| 140 |
+
Metrics:
|
| 141 |
+
box AP: 51.2
|
| 142 |
+
Weights: https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_n-v61_fast_1xb64-50e_voc/yolov5_n-v61_fast_1xb64-50e_voc_20221017_234254-f1493430.pth
|
| 143 |
+
- Name: yolov5_s-v61_fast_1xb64-50e_voc
|
| 144 |
+
In Collection: YOLOv5_VOC
|
| 145 |
+
Config: configs/yolov5/voc/yolov5_s-v61_fast_1xb64-50e_voc.py
|
| 146 |
+
Metadata:
|
| 147 |
+
Training Memory (GB): 6.5
|
| 148 |
+
Epochs: 50
|
| 149 |
+
Results:
|
| 150 |
+
- Task: Object Detection
|
| 151 |
+
Dataset: COCO
|
| 152 |
+
Metrics:
|
| 153 |
+
box AP: 62.7
|
| 154 |
+
Weights: https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_s-v61_fast_1xb64-50e_voc/yolov5_s-v61_fast_1xb64-50e_voc_20221017_234156-0009b33e.pth
|
| 155 |
+
- Name: yolov5_m-v61_fast_1xb64-50e_voc
|
| 156 |
+
In Collection: YOLOv5_VOC
|
| 157 |
+
Config: configs/yolov5/voc/yolov5_m-v61_fast_1xb64-50e_voc.py
|
| 158 |
+
Metadata:
|
| 159 |
+
Training Memory (GB): 12.0
|
| 160 |
+
Epochs: 50
|
| 161 |
+
Results:
|
| 162 |
+
- Task: Object Detection
|
| 163 |
+
Dataset: COCO
|
| 164 |
+
Metrics:
|
| 165 |
+
box AP: 70.1
|
| 166 |
+
Weights: https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_m-v61_fast_1xb64-50e_voc/yolov5_m-v61_fast_1xb64-50e_voc_20221017_114138-815c143a.pth
|
| 167 |
+
- Name: yolov5_l-v61_fast_1xb32-50e_voc
|
| 168 |
+
In Collection: YOLOv5_VOC
|
| 169 |
+
Config: configs/yolov5/voc/yolov5_l-v61_fast_1xb32-50e_voc.py
|
| 170 |
+
Metadata:
|
| 171 |
+
Training Memory (GB): 10.0
|
| 172 |
+
Epochs: 50
|
| 173 |
+
Results:
|
| 174 |
+
- Task: Object Detection
|
| 175 |
+
Dataset: COCO
|
| 176 |
+
Metrics:
|
| 177 |
+
box AP: 73.1
|
| 178 |
+
Weights: https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_l-v61_fast_1xb32-50e_voc/yolov5_l-v61_fast_1xb32-50e_voc_20221017_045500-edc7e0d8.pth
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/voc/__init__.py
ADDED
|
File without changes
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/voc/yolov5_l-v61_fast_1xb32-50e_voc.py
ADDED
|
@@ -0,0 +1,25 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = './yolov5_s-v61_fast_1xb64-50e_voc.py'
|
| 2 |
+
|
| 3 |
+
deepen_factor = 1.0
|
| 4 |
+
widen_factor = 1.0
|
| 5 |
+
train_batch_size_per_gpu = 32
|
| 6 |
+
train_num_workers = 8
|
| 7 |
+
|
| 8 |
+
load_from = 'https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_l-v61_syncbn_fast_8xb16-300e_coco/yolov5_l-v61_syncbn_fast_8xb16-300e_coco_20220917_031007-096ef0eb.pth' # noqa
|
| 9 |
+
|
| 10 |
+
model = dict(
|
| 11 |
+
backbone=dict(
|
| 12 |
+
deepen_factor=deepen_factor,
|
| 13 |
+
widen_factor=widen_factor,
|
| 14 |
+
),
|
| 15 |
+
neck=dict(
|
| 16 |
+
deepen_factor=deepen_factor,
|
| 17 |
+
widen_factor=widen_factor,
|
| 18 |
+
),
|
| 19 |
+
bbox_head=dict(head_module=dict(widen_factor=widen_factor)))
|
| 20 |
+
|
| 21 |
+
train_dataloader = dict(
|
| 22 |
+
batch_size=train_batch_size_per_gpu, num_workers=train_num_workers)
|
| 23 |
+
|
| 24 |
+
optim_wrapper = dict(
|
| 25 |
+
optimizer=dict(batch_size_per_gpu=train_batch_size_per_gpu))
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/voc/yolov5_m-v61_fast_1xb64-50e_voc.py
ADDED
|
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = './yolov5_s-v61_fast_1xb64-50e_voc.py'
|
| 2 |
+
|
| 3 |
+
deepen_factor = 0.67
|
| 4 |
+
widen_factor = 0.75
|
| 5 |
+
|
| 6 |
+
load_from = 'https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_m-v61_syncbn_fast_8xb16-300e_coco/yolov5_m-v61_syncbn_fast_8xb16-300e_coco_20220917_204944-516a710f.pth' # noqa
|
| 7 |
+
|
| 8 |
+
model = dict(
|
| 9 |
+
backbone=dict(
|
| 10 |
+
deepen_factor=deepen_factor,
|
| 11 |
+
widen_factor=widen_factor,
|
| 12 |
+
),
|
| 13 |
+
neck=dict(
|
| 14 |
+
deepen_factor=deepen_factor,
|
| 15 |
+
widen_factor=widen_factor,
|
| 16 |
+
),
|
| 17 |
+
bbox_head=dict(head_module=dict(widen_factor=widen_factor)))
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/voc/yolov5_n-v61_fast_1xb64-50e_voc.py
ADDED
|
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = './yolov5_s-v61_fast_1xb64-50e_voc.py'
|
| 2 |
+
|
| 3 |
+
deepen_factor = 0.33
|
| 4 |
+
widen_factor = 0.25
|
| 5 |
+
|
| 6 |
+
load_from = 'https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_n-v61_syncbn_fast_8xb16-300e_coco/yolov5_n-v61_syncbn_fast_8xb16-300e_coco_20220919_090739-b804c1ad.pth' # noqa
|
| 7 |
+
|
| 8 |
+
model = dict(
|
| 9 |
+
backbone=dict(
|
| 10 |
+
deepen_factor=deepen_factor,
|
| 11 |
+
widen_factor=widen_factor,
|
| 12 |
+
),
|
| 13 |
+
neck=dict(
|
| 14 |
+
deepen_factor=deepen_factor,
|
| 15 |
+
widen_factor=widen_factor,
|
| 16 |
+
),
|
| 17 |
+
bbox_head=dict(head_module=dict(widen_factor=widen_factor)))
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/voc/yolov5_s-v61_fast_1xb64-50e_voc.py
ADDED
|
@@ -0,0 +1,270 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = '../yolov5_s-v61_syncbn_fast_8xb16-300e_coco.py'
|
| 2 |
+
|
| 3 |
+
# dataset settings
|
| 4 |
+
data_root = 'data/VOCdevkit/'
|
| 5 |
+
dataset_type = 'YOLOv5VOCDataset'
|
| 6 |
+
|
| 7 |
+
# parameters that often need to be modified
|
| 8 |
+
num_classes = 20
|
| 9 |
+
img_scale = (512, 512) # width, height
|
| 10 |
+
max_epochs = 50
|
| 11 |
+
train_batch_size_per_gpu = 64
|
| 12 |
+
train_num_workers = 8
|
| 13 |
+
val_batch_size_per_gpu = 1
|
| 14 |
+
val_num_workers = 2
|
| 15 |
+
|
| 16 |
+
# persistent_workers must be False if num_workers is 0.
|
| 17 |
+
persistent_workers = True
|
| 18 |
+
|
| 19 |
+
lr_factor = 0.15135
|
| 20 |
+
affine_scale = 0.75544
|
| 21 |
+
|
| 22 |
+
# only on Val
|
| 23 |
+
batch_shapes_cfg = dict(img_size=img_scale[0])
|
| 24 |
+
|
| 25 |
+
anchors = [[(26, 44), (67, 57), (61, 130)], [(121, 118), (120, 239),
|
| 26 |
+
(206, 182)],
|
| 27 |
+
[(376, 161), (234, 324), (428, 322)]]
|
| 28 |
+
num_det_layers = 3
|
| 29 |
+
|
| 30 |
+
load_from = 'https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_s-v61_syncbn_fast_8xb16-300e_coco/yolov5_s-v61_syncbn_fast_8xb16-300e_coco_20220918_084700-86e02187.pth' # noqa
|
| 31 |
+
|
| 32 |
+
tta_img_scales = [img_scale, (416, 416), (640, 640)]
|
| 33 |
+
|
| 34 |
+
# Hyperparameter reference from:
|
| 35 |
+
# https://github.com/ultralytics/yolov5/blob/master/data/hyps/hyp.VOC.yaml
|
| 36 |
+
model = dict(
|
| 37 |
+
bbox_head=dict(
|
| 38 |
+
head_module=dict(num_classes=num_classes),
|
| 39 |
+
prior_generator=dict(base_sizes=anchors),
|
| 40 |
+
loss_cls=dict(
|
| 41 |
+
loss_weight=0.21638 * (num_classes / 80 * 3 / num_det_layers),
|
| 42 |
+
class_weight=0.5),
|
| 43 |
+
loss_bbox=dict(loss_weight=0.02 * (3 / num_det_layers)),
|
| 44 |
+
loss_obj=dict(
|
| 45 |
+
loss_weight=0.51728 *
|
| 46 |
+
((img_scale[0] / 640)**2 * 3 / num_det_layers),
|
| 47 |
+
class_weight=0.67198),
|
| 48 |
+
# Different from COCO
|
| 49 |
+
prior_match_thr=3.3744),
|
| 50 |
+
test_cfg=dict(nms=dict(iou_threshold=0.6)))
|
| 51 |
+
|
| 52 |
+
albu_train_transforms = _base_.albu_train_transforms
|
| 53 |
+
pre_transform = _base_.pre_transform
|
| 54 |
+
|
| 55 |
+
with_mosiac_pipeline = [
|
| 56 |
+
dict(
|
| 57 |
+
type='Mosaic',
|
| 58 |
+
img_scale=img_scale,
|
| 59 |
+
pad_val=114.0,
|
| 60 |
+
pre_transform=pre_transform),
|
| 61 |
+
dict(
|
| 62 |
+
type='YOLOv5RandomAffine',
|
| 63 |
+
max_rotate_degree=0.0,
|
| 64 |
+
max_translate_ratio=0.04591,
|
| 65 |
+
max_shear_degree=0.0,
|
| 66 |
+
scaling_ratio_range=(1 - affine_scale, 1 + affine_scale),
|
| 67 |
+
# img_scale is (width, height)
|
| 68 |
+
border=(-img_scale[0] // 2, -img_scale[1] // 2),
|
| 69 |
+
border_val=(114, 114, 114)),
|
| 70 |
+
dict(
|
| 71 |
+
type='YOLOv5MixUp',
|
| 72 |
+
prob=0.04266,
|
| 73 |
+
pre_transform=[
|
| 74 |
+
*pre_transform,
|
| 75 |
+
dict(
|
| 76 |
+
type='Mosaic',
|
| 77 |
+
img_scale=img_scale,
|
| 78 |
+
pad_val=114.0,
|
| 79 |
+
pre_transform=pre_transform),
|
| 80 |
+
dict(
|
| 81 |
+
type='YOLOv5RandomAffine',
|
| 82 |
+
max_rotate_degree=0.0,
|
| 83 |
+
max_translate_ratio=0.04591,
|
| 84 |
+
max_shear_degree=0.0,
|
| 85 |
+
scaling_ratio_range=(1 - affine_scale, 1 + affine_scale),
|
| 86 |
+
# img_scale is (width, height)
|
| 87 |
+
border=(-img_scale[0] // 2, -img_scale[1] // 2),
|
| 88 |
+
border_val=(114, 114, 114))
|
| 89 |
+
])
|
| 90 |
+
]
|
| 91 |
+
|
| 92 |
+
without_mosaic_pipeline = [
|
| 93 |
+
dict(
|
| 94 |
+
type='YOLOv5RandomAffine',
|
| 95 |
+
max_rotate_degree=0.0,
|
| 96 |
+
max_translate_ratio=0.04591,
|
| 97 |
+
max_shear_degree=0.0,
|
| 98 |
+
scaling_ratio_range=(1 - affine_scale, 1 + affine_scale),
|
| 99 |
+
border=(0, 0),
|
| 100 |
+
border_val=(114, 114, 114)),
|
| 101 |
+
dict(
|
| 102 |
+
type='LetterResize',
|
| 103 |
+
scale=img_scale,
|
| 104 |
+
allow_scale_up=True,
|
| 105 |
+
pad_val=dict(img=114))
|
| 106 |
+
]
|
| 107 |
+
|
| 108 |
+
# Because the border parameter is inconsistent when
|
| 109 |
+
# using mosaic or not, `RandomChoice` is used here.
|
| 110 |
+
randchoice_mosaic_pipeline = dict(
|
| 111 |
+
type='RandomChoice',
|
| 112 |
+
transforms=[with_mosiac_pipeline, without_mosaic_pipeline],
|
| 113 |
+
prob=[0.85834, 0.14166])
|
| 114 |
+
|
| 115 |
+
train_pipeline = [
|
| 116 |
+
*pre_transform, randchoice_mosaic_pipeline,
|
| 117 |
+
dict(
|
| 118 |
+
type='mmdet.Albu',
|
| 119 |
+
transforms=albu_train_transforms,
|
| 120 |
+
bbox_params=dict(
|
| 121 |
+
type='BboxParams',
|
| 122 |
+
format='pascal_voc',
|
| 123 |
+
label_fields=['gt_bboxes_labels', 'gt_ignore_flags']),
|
| 124 |
+
keymap={
|
| 125 |
+
'img': 'image',
|
| 126 |
+
'gt_bboxes': 'bboxes'
|
| 127 |
+
}),
|
| 128 |
+
dict(
|
| 129 |
+
type='YOLOv5HSVRandomAug',
|
| 130 |
+
hue_delta=0.01041,
|
| 131 |
+
saturation_delta=0.54703,
|
| 132 |
+
value_delta=0.27739),
|
| 133 |
+
dict(type='mmdet.RandomFlip', prob=0.5),
|
| 134 |
+
dict(
|
| 135 |
+
type='mmdet.PackDetInputs',
|
| 136 |
+
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape', 'flip',
|
| 137 |
+
'flip_direction'))
|
| 138 |
+
]
|
| 139 |
+
|
| 140 |
+
train_dataloader = dict(
|
| 141 |
+
_delete_=True,
|
| 142 |
+
batch_size=train_batch_size_per_gpu,
|
| 143 |
+
num_workers=train_num_workers,
|
| 144 |
+
persistent_workers=persistent_workers,
|
| 145 |
+
pin_memory=True,
|
| 146 |
+
sampler=dict(type='DefaultSampler', shuffle=True),
|
| 147 |
+
dataset=dict(
|
| 148 |
+
type='ConcatDataset',
|
| 149 |
+
datasets=[
|
| 150 |
+
dict(
|
| 151 |
+
type=dataset_type,
|
| 152 |
+
data_root=data_root,
|
| 153 |
+
ann_file='VOC2007/ImageSets/Main/trainval.txt',
|
| 154 |
+
data_prefix=dict(sub_data_root='VOC2007/'),
|
| 155 |
+
filter_cfg=dict(filter_empty_gt=False, min_size=32),
|
| 156 |
+
pipeline=train_pipeline),
|
| 157 |
+
dict(
|
| 158 |
+
type=dataset_type,
|
| 159 |
+
data_root=data_root,
|
| 160 |
+
ann_file='VOC2012/ImageSets/Main/trainval.txt',
|
| 161 |
+
data_prefix=dict(sub_data_root='VOC2012/'),
|
| 162 |
+
filter_cfg=dict(filter_empty_gt=False, min_size=32),
|
| 163 |
+
pipeline=train_pipeline)
|
| 164 |
+
],
|
| 165 |
+
# Use ignore_keys to avoid judging metainfo is
|
| 166 |
+
# not equal in `ConcatDataset`.
|
| 167 |
+
ignore_keys='dataset_type'),
|
| 168 |
+
collate_fn=dict(type='yolov5_collate'))
|
| 169 |
+
|
| 170 |
+
test_pipeline = [
|
| 171 |
+
dict(type='LoadImageFromFile', file_client_args=_base_.file_client_args),
|
| 172 |
+
dict(type='YOLOv5KeepRatioResize', scale=img_scale),
|
| 173 |
+
dict(
|
| 174 |
+
type='LetterResize',
|
| 175 |
+
scale=img_scale,
|
| 176 |
+
allow_scale_up=False,
|
| 177 |
+
pad_val=dict(img=114)),
|
| 178 |
+
dict(type='LoadAnnotations', with_bbox=True, _scope_='mmdet'),
|
| 179 |
+
dict(
|
| 180 |
+
type='mmdet.PackDetInputs',
|
| 181 |
+
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
|
| 182 |
+
'scale_factor', 'pad_param'))
|
| 183 |
+
]
|
| 184 |
+
|
| 185 |
+
val_dataloader = dict(
|
| 186 |
+
batch_size=val_batch_size_per_gpu,
|
| 187 |
+
num_workers=val_num_workers,
|
| 188 |
+
persistent_workers=persistent_workers,
|
| 189 |
+
pin_memory=True,
|
| 190 |
+
drop_last=False,
|
| 191 |
+
sampler=dict(type='DefaultSampler', shuffle=False),
|
| 192 |
+
dataset=dict(
|
| 193 |
+
type=dataset_type,
|
| 194 |
+
data_root=data_root,
|
| 195 |
+
ann_file='VOC2007/ImageSets/Main/test.txt',
|
| 196 |
+
data_prefix=dict(sub_data_root='VOC2007/'),
|
| 197 |
+
test_mode=True,
|
| 198 |
+
pipeline=test_pipeline,
|
| 199 |
+
batch_shapes_cfg=batch_shapes_cfg))
|
| 200 |
+
|
| 201 |
+
test_dataloader = val_dataloader
|
| 202 |
+
|
| 203 |
+
param_scheduler = None
|
| 204 |
+
optim_wrapper = dict(
|
| 205 |
+
optimizer=dict(
|
| 206 |
+
lr=0.00334,
|
| 207 |
+
momentum=0.74832,
|
| 208 |
+
weight_decay=0.00025,
|
| 209 |
+
batch_size_per_gpu=train_batch_size_per_gpu))
|
| 210 |
+
|
| 211 |
+
default_hooks = dict(
|
| 212 |
+
param_scheduler=dict(
|
| 213 |
+
lr_factor=lr_factor,
|
| 214 |
+
max_epochs=max_epochs,
|
| 215 |
+
warmup_epochs=3.3835,
|
| 216 |
+
warmup_momentum=0.59462,
|
| 217 |
+
warmup_bias_lr=0.18657))
|
| 218 |
+
|
| 219 |
+
custom_hooks = [
|
| 220 |
+
dict(
|
| 221 |
+
type='EMAHook',
|
| 222 |
+
ema_type='ExpMomentumEMA',
|
| 223 |
+
momentum=0.0001,
|
| 224 |
+
update_buffers=True,
|
| 225 |
+
# To load COCO pretrained model, need to set `strict_load=False`
|
| 226 |
+
strict_load=False,
|
| 227 |
+
priority=49)
|
| 228 |
+
]
|
| 229 |
+
|
| 230 |
+
# TODO: Support using coco metric in voc dataset
|
| 231 |
+
val_evaluator = dict(
|
| 232 |
+
_delete_=True, type='mmdet.VOCMetric', metric='mAP', eval_mode='area')
|
| 233 |
+
|
| 234 |
+
test_evaluator = val_evaluator
|
| 235 |
+
|
| 236 |
+
train_cfg = dict(max_epochs=max_epochs)
|
| 237 |
+
|
| 238 |
+
# Config for Test Time Augmentation. (TTA)
|
| 239 |
+
_multiscale_resize_transforms = [
|
| 240 |
+
dict(
|
| 241 |
+
type='Compose',
|
| 242 |
+
transforms=[
|
| 243 |
+
dict(type='YOLOv5KeepRatioResize', scale=s),
|
| 244 |
+
dict(
|
| 245 |
+
type='LetterResize',
|
| 246 |
+
scale=s,
|
| 247 |
+
allow_scale_up=False,
|
| 248 |
+
pad_val=dict(img=114))
|
| 249 |
+
]) for s in tta_img_scales
|
| 250 |
+
]
|
| 251 |
+
|
| 252 |
+
tta_pipeline = [
|
| 253 |
+
dict(type='LoadImageFromFile', file_client_args=_base_.file_client_args),
|
| 254 |
+
dict(
|
| 255 |
+
type='TestTimeAug',
|
| 256 |
+
transforms=[
|
| 257 |
+
_multiscale_resize_transforms,
|
| 258 |
+
[
|
| 259 |
+
dict(type='mmdet.RandomFlip', prob=1.),
|
| 260 |
+
dict(type='mmdet.RandomFlip', prob=0.)
|
| 261 |
+
], [dict(type='mmdet.LoadAnnotations', with_bbox=True)],
|
| 262 |
+
[
|
| 263 |
+
dict(
|
| 264 |
+
type='mmdet.PackDetInputs',
|
| 265 |
+
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
|
| 266 |
+
'scale_factor', 'pad_param', 'flip',
|
| 267 |
+
'flip_direction'))
|
| 268 |
+
]
|
| 269 |
+
])
|
| 270 |
+
]
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/voc/yolov5_x-v61_fast_1xb32-50e_voc.py
ADDED
|
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = './yolov5_s-v61_fast_1xb64-50e_voc.py'
|
| 2 |
+
|
| 3 |
+
deepen_factor = 1.33
|
| 4 |
+
widen_factor = 1.25
|
| 5 |
+
train_batch_size_per_gpu = 32
|
| 6 |
+
train_num_workers = 8
|
| 7 |
+
|
| 8 |
+
# TODO: need to add pretrained_model
|
| 9 |
+
load_from = None
|
| 10 |
+
|
| 11 |
+
model = dict(
|
| 12 |
+
backbone=dict(
|
| 13 |
+
deepen_factor=deepen_factor,
|
| 14 |
+
widen_factor=widen_factor,
|
| 15 |
+
),
|
| 16 |
+
neck=dict(
|
| 17 |
+
deepen_factor=deepen_factor,
|
| 18 |
+
widen_factor=widen_factor,
|
| 19 |
+
),
|
| 20 |
+
bbox_head=dict(head_module=dict(widen_factor=widen_factor)))
|
| 21 |
+
|
| 22 |
+
train_dataloader = dict(
|
| 23 |
+
batch_size=train_batch_size_per_gpu, num_workers=train_num_workers)
|
| 24 |
+
|
| 25 |
+
optim_wrapper = dict(
|
| 26 |
+
optimizer=dict(batch_size_per_gpu=train_batch_size_per_gpu))
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_l-p6-v62_syncbn_fast_8xb16-300e_coco.py
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = './yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco.py'
|
| 2 |
+
|
| 3 |
+
deepen_factor = 1.0
|
| 4 |
+
widen_factor = 1.0
|
| 5 |
+
|
| 6 |
+
model = dict(
|
| 7 |
+
backbone=dict(
|
| 8 |
+
deepen_factor=deepen_factor,
|
| 9 |
+
widen_factor=widen_factor,
|
| 10 |
+
),
|
| 11 |
+
neck=dict(
|
| 12 |
+
deepen_factor=deepen_factor,
|
| 13 |
+
widen_factor=widen_factor,
|
| 14 |
+
),
|
| 15 |
+
bbox_head=dict(head_module=dict(widen_factor=widen_factor)))
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_l-v61_syncbn_fast_8xb16-300e_coco.py
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = './yolov5_m-v61_syncbn_fast_8xb16-300e_coco.py'
|
| 2 |
+
|
| 3 |
+
deepen_factor = 1.0
|
| 4 |
+
widen_factor = 1.0
|
| 5 |
+
|
| 6 |
+
model = dict(
|
| 7 |
+
backbone=dict(
|
| 8 |
+
deepen_factor=deepen_factor,
|
| 9 |
+
widen_factor=widen_factor,
|
| 10 |
+
),
|
| 11 |
+
neck=dict(
|
| 12 |
+
deepen_factor=deepen_factor,
|
| 13 |
+
widen_factor=widen_factor,
|
| 14 |
+
),
|
| 15 |
+
bbox_head=dict(head_module=dict(widen_factor=widen_factor)))
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco.py
ADDED
|
@@ -0,0 +1,79 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = './yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco.py'
|
| 2 |
+
|
| 3 |
+
# ========================modified parameters======================
|
| 4 |
+
deepen_factor = 0.67
|
| 5 |
+
widen_factor = 0.75
|
| 6 |
+
lr_factor = 0.1
|
| 7 |
+
affine_scale = 0.9
|
| 8 |
+
loss_cls_weight = 0.3
|
| 9 |
+
loss_obj_weight = 0.7
|
| 10 |
+
mixup_prob = 0.1
|
| 11 |
+
|
| 12 |
+
# =======================Unmodified in most cases==================
|
| 13 |
+
num_classes = _base_.num_classes
|
| 14 |
+
num_det_layers = _base_.num_det_layers
|
| 15 |
+
img_scale = _base_.img_scale
|
| 16 |
+
|
| 17 |
+
model = dict(
|
| 18 |
+
backbone=dict(
|
| 19 |
+
deepen_factor=deepen_factor,
|
| 20 |
+
widen_factor=widen_factor,
|
| 21 |
+
),
|
| 22 |
+
neck=dict(
|
| 23 |
+
deepen_factor=deepen_factor,
|
| 24 |
+
widen_factor=widen_factor,
|
| 25 |
+
),
|
| 26 |
+
bbox_head=dict(
|
| 27 |
+
head_module=dict(widen_factor=widen_factor),
|
| 28 |
+
loss_cls=dict(loss_weight=loss_cls_weight *
|
| 29 |
+
(num_classes / 80 * 3 / num_det_layers)),
|
| 30 |
+
loss_obj=dict(loss_weight=loss_obj_weight *
|
| 31 |
+
((img_scale[0] / 640)**2 * 3 / num_det_layers))))
|
| 32 |
+
|
| 33 |
+
pre_transform = _base_.pre_transform
|
| 34 |
+
albu_train_transforms = _base_.albu_train_transforms
|
| 35 |
+
|
| 36 |
+
mosaic_affine_pipeline = [
|
| 37 |
+
dict(
|
| 38 |
+
type='Mosaic',
|
| 39 |
+
img_scale=img_scale,
|
| 40 |
+
pad_val=114.0,
|
| 41 |
+
pre_transform=pre_transform),
|
| 42 |
+
dict(
|
| 43 |
+
type='YOLOv5RandomAffine',
|
| 44 |
+
max_rotate_degree=0.0,
|
| 45 |
+
max_shear_degree=0.0,
|
| 46 |
+
scaling_ratio_range=(1 - affine_scale, 1 + affine_scale),
|
| 47 |
+
# img_scale is (width, height)
|
| 48 |
+
border=(-img_scale[0] // 2, -img_scale[1] // 2),
|
| 49 |
+
border_val=(114, 114, 114))
|
| 50 |
+
]
|
| 51 |
+
|
| 52 |
+
# enable mixup
|
| 53 |
+
train_pipeline = [
|
| 54 |
+
*pre_transform, *mosaic_affine_pipeline,
|
| 55 |
+
dict(
|
| 56 |
+
type='YOLOv5MixUp',
|
| 57 |
+
prob=mixup_prob,
|
| 58 |
+
pre_transform=[*pre_transform, *mosaic_affine_pipeline]),
|
| 59 |
+
dict(
|
| 60 |
+
type='mmdet.Albu',
|
| 61 |
+
transforms=albu_train_transforms,
|
| 62 |
+
bbox_params=dict(
|
| 63 |
+
type='BboxParams',
|
| 64 |
+
format='pascal_voc',
|
| 65 |
+
label_fields=['gt_bboxes_labels', 'gt_ignore_flags']),
|
| 66 |
+
keymap={
|
| 67 |
+
'img': 'image',
|
| 68 |
+
'gt_bboxes': 'bboxes'
|
| 69 |
+
}),
|
| 70 |
+
dict(type='YOLOv5HSVRandomAug'),
|
| 71 |
+
dict(type='mmdet.RandomFlip', prob=0.5),
|
| 72 |
+
dict(
|
| 73 |
+
type='mmdet.PackDetInputs',
|
| 74 |
+
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape', 'flip',
|
| 75 |
+
'flip_direction'))
|
| 76 |
+
]
|
| 77 |
+
|
| 78 |
+
train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
|
| 79 |
+
default_hooks = dict(param_scheduler=dict(lr_factor=lr_factor))
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_m-v61_syncbn_fast_8xb16-300e_coco.py
ADDED
|
@@ -0,0 +1,79 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = './yolov5_s-v61_syncbn_fast_8xb16-300e_coco.py'
|
| 2 |
+
|
| 3 |
+
# ========================modified parameters======================
|
| 4 |
+
deepen_factor = 0.67
|
| 5 |
+
widen_factor = 0.75
|
| 6 |
+
lr_factor = 0.1
|
| 7 |
+
affine_scale = 0.9
|
| 8 |
+
loss_cls_weight = 0.3
|
| 9 |
+
loss_obj_weight = 0.7
|
| 10 |
+
mixup_prob = 0.1
|
| 11 |
+
|
| 12 |
+
# =======================Unmodified in most cases==================
|
| 13 |
+
num_classes = _base_.num_classes
|
| 14 |
+
num_det_layers = _base_.num_det_layers
|
| 15 |
+
img_scale = _base_.img_scale
|
| 16 |
+
|
| 17 |
+
model = dict(
|
| 18 |
+
backbone=dict(
|
| 19 |
+
deepen_factor=deepen_factor,
|
| 20 |
+
widen_factor=widen_factor,
|
| 21 |
+
),
|
| 22 |
+
neck=dict(
|
| 23 |
+
deepen_factor=deepen_factor,
|
| 24 |
+
widen_factor=widen_factor,
|
| 25 |
+
),
|
| 26 |
+
bbox_head=dict(
|
| 27 |
+
head_module=dict(widen_factor=widen_factor),
|
| 28 |
+
loss_cls=dict(loss_weight=loss_cls_weight *
|
| 29 |
+
(num_classes / 80 * 3 / num_det_layers)),
|
| 30 |
+
loss_obj=dict(loss_weight=loss_obj_weight *
|
| 31 |
+
((img_scale[0] / 640)**2 * 3 / num_det_layers))))
|
| 32 |
+
|
| 33 |
+
pre_transform = _base_.pre_transform
|
| 34 |
+
albu_train_transforms = _base_.albu_train_transforms
|
| 35 |
+
|
| 36 |
+
mosaic_affine_pipeline = [
|
| 37 |
+
dict(
|
| 38 |
+
type='Mosaic',
|
| 39 |
+
img_scale=img_scale,
|
| 40 |
+
pad_val=114.0,
|
| 41 |
+
pre_transform=pre_transform),
|
| 42 |
+
dict(
|
| 43 |
+
type='YOLOv5RandomAffine',
|
| 44 |
+
max_rotate_degree=0.0,
|
| 45 |
+
max_shear_degree=0.0,
|
| 46 |
+
scaling_ratio_range=(1 - affine_scale, 1 + affine_scale),
|
| 47 |
+
# img_scale is (width, height)
|
| 48 |
+
border=(-img_scale[0] // 2, -img_scale[1] // 2),
|
| 49 |
+
border_val=(114, 114, 114))
|
| 50 |
+
]
|
| 51 |
+
|
| 52 |
+
# enable mixup
|
| 53 |
+
train_pipeline = [
|
| 54 |
+
*pre_transform, *mosaic_affine_pipeline,
|
| 55 |
+
dict(
|
| 56 |
+
type='YOLOv5MixUp',
|
| 57 |
+
prob=mixup_prob,
|
| 58 |
+
pre_transform=[*pre_transform, *mosaic_affine_pipeline]),
|
| 59 |
+
dict(
|
| 60 |
+
type='mmdet.Albu',
|
| 61 |
+
transforms=albu_train_transforms,
|
| 62 |
+
bbox_params=dict(
|
| 63 |
+
type='BboxParams',
|
| 64 |
+
format='pascal_voc',
|
| 65 |
+
label_fields=['gt_bboxes_labels', 'gt_ignore_flags']),
|
| 66 |
+
keymap={
|
| 67 |
+
'img': 'image',
|
| 68 |
+
'gt_bboxes': 'bboxes'
|
| 69 |
+
}),
|
| 70 |
+
dict(type='YOLOv5HSVRandomAug'),
|
| 71 |
+
dict(type='mmdet.RandomFlip', prob=0.5),
|
| 72 |
+
dict(
|
| 73 |
+
type='mmdet.PackDetInputs',
|
| 74 |
+
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape', 'flip',
|
| 75 |
+
'flip_direction'))
|
| 76 |
+
]
|
| 77 |
+
|
| 78 |
+
train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
|
| 79 |
+
default_hooks = dict(param_scheduler=dict(lr_factor=lr_factor))
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_n-p6-v62_syncbn_fast_8xb16-300e_coco.py
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = 'yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco.py'
|
| 2 |
+
|
| 3 |
+
deepen_factor = 0.33
|
| 4 |
+
widen_factor = 0.25
|
| 5 |
+
|
| 6 |
+
model = dict(
|
| 7 |
+
backbone=dict(
|
| 8 |
+
deepen_factor=deepen_factor,
|
| 9 |
+
widen_factor=widen_factor,
|
| 10 |
+
),
|
| 11 |
+
neck=dict(
|
| 12 |
+
deepen_factor=deepen_factor,
|
| 13 |
+
widen_factor=widen_factor,
|
| 14 |
+
),
|
| 15 |
+
bbox_head=dict(head_module=dict(widen_factor=widen_factor)))
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_n-v61_syncbn_fast_8xb16-300e_coco.py
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = './yolov5_s-v61_syncbn_fast_8xb16-300e_coco.py'
|
| 2 |
+
|
| 3 |
+
deepen_factor = 0.33
|
| 4 |
+
widen_factor = 0.25
|
| 5 |
+
|
| 6 |
+
model = dict(
|
| 7 |
+
backbone=dict(
|
| 8 |
+
deepen_factor=deepen_factor,
|
| 9 |
+
widen_factor=widen_factor,
|
| 10 |
+
),
|
| 11 |
+
neck=dict(
|
| 12 |
+
deepen_factor=deepen_factor,
|
| 13 |
+
widen_factor=widen_factor,
|
| 14 |
+
),
|
| 15 |
+
bbox_head=dict(head_module=dict(widen_factor=widen_factor)))
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_s-p6-v62_syncbn_fast_8xb16-300e_coco.py
ADDED
|
@@ -0,0 +1,138 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = 'yolov5_s-v61_syncbn_fast_8xb16-300e_coco.py'
|
| 2 |
+
|
| 3 |
+
# ========================modified parameters======================
|
| 4 |
+
img_scale = (1280, 1280) # width, height
|
| 5 |
+
num_classes = 80 # Number of classes for classification
|
| 6 |
+
# Config of batch shapes. Only on val.
|
| 7 |
+
# It means not used if batch_shapes_cfg is None.
|
| 8 |
+
batch_shapes_cfg = dict(
|
| 9 |
+
img_size=img_scale[0],
|
| 10 |
+
# The image scale of padding should be divided by pad_size_divisor
|
| 11 |
+
size_divisor=64)
|
| 12 |
+
# Basic size of multi-scale prior box
|
| 13 |
+
anchors = [
|
| 14 |
+
[(19, 27), (44, 40), (38, 94)], # P3/8
|
| 15 |
+
[(96, 68), (86, 152), (180, 137)], # P4/16
|
| 16 |
+
[(140, 301), (303, 264), (238, 542)], # P5/32
|
| 17 |
+
[(436, 615), (739, 380), (925, 792)] # P6/64
|
| 18 |
+
]
|
| 19 |
+
# Strides of multi-scale prior box
|
| 20 |
+
strides = [8, 16, 32, 64]
|
| 21 |
+
num_det_layers = 4 # The number of model output scales
|
| 22 |
+
loss_cls_weight = 0.5
|
| 23 |
+
loss_bbox_weight = 0.05
|
| 24 |
+
loss_obj_weight = 1.0
|
| 25 |
+
# The obj loss weights of the three output layers
|
| 26 |
+
obj_level_weights = [4.0, 1.0, 0.25, 0.06]
|
| 27 |
+
affine_scale = 0.5 # YOLOv5RandomAffine scaling ratio
|
| 28 |
+
|
| 29 |
+
tta_img_scales = [(1280, 1280), (1024, 1024), (1536, 1536)]
|
| 30 |
+
# =======================Unmodified in most cases==================
|
| 31 |
+
model = dict(
|
| 32 |
+
backbone=dict(arch='P6', out_indices=(2, 3, 4, 5)),
|
| 33 |
+
neck=dict(
|
| 34 |
+
in_channels=[256, 512, 768, 1024], out_channels=[256, 512, 768, 1024]),
|
| 35 |
+
bbox_head=dict(
|
| 36 |
+
head_module=dict(
|
| 37 |
+
in_channels=[256, 512, 768, 1024], featmap_strides=strides),
|
| 38 |
+
prior_generator=dict(base_sizes=anchors, strides=strides),
|
| 39 |
+
# scaled based on number of detection layers
|
| 40 |
+
loss_cls=dict(loss_weight=loss_cls_weight *
|
| 41 |
+
(num_classes / 80 * 3 / num_det_layers)),
|
| 42 |
+
loss_bbox=dict(loss_weight=loss_bbox_weight * (3 / num_det_layers)),
|
| 43 |
+
loss_obj=dict(loss_weight=loss_obj_weight *
|
| 44 |
+
((img_scale[0] / 640)**2 * 3 / num_det_layers)),
|
| 45 |
+
obj_level_weights=obj_level_weights))
|
| 46 |
+
|
| 47 |
+
pre_transform = _base_.pre_transform
|
| 48 |
+
albu_train_transforms = _base_.albu_train_transforms
|
| 49 |
+
|
| 50 |
+
train_pipeline = [
|
| 51 |
+
*pre_transform,
|
| 52 |
+
dict(
|
| 53 |
+
type='Mosaic',
|
| 54 |
+
img_scale=img_scale,
|
| 55 |
+
pad_val=114.0,
|
| 56 |
+
pre_transform=pre_transform),
|
| 57 |
+
dict(
|
| 58 |
+
type='YOLOv5RandomAffine',
|
| 59 |
+
max_rotate_degree=0.0,
|
| 60 |
+
max_shear_degree=0.0,
|
| 61 |
+
scaling_ratio_range=(1 - affine_scale, 1 + affine_scale),
|
| 62 |
+
# img_scale is (width, height)
|
| 63 |
+
border=(-img_scale[0] // 2, -img_scale[1] // 2),
|
| 64 |
+
border_val=(114, 114, 114)),
|
| 65 |
+
dict(
|
| 66 |
+
type='mmdet.Albu',
|
| 67 |
+
transforms=albu_train_transforms,
|
| 68 |
+
bbox_params=dict(
|
| 69 |
+
type='BboxParams',
|
| 70 |
+
format='pascal_voc',
|
| 71 |
+
label_fields=['gt_bboxes_labels', 'gt_ignore_flags']),
|
| 72 |
+
keymap={
|
| 73 |
+
'img': 'image',
|
| 74 |
+
'gt_bboxes': 'bboxes'
|
| 75 |
+
}),
|
| 76 |
+
dict(type='YOLOv5HSVRandomAug'),
|
| 77 |
+
dict(type='mmdet.RandomFlip', prob=0.5),
|
| 78 |
+
dict(
|
| 79 |
+
type='mmdet.PackDetInputs',
|
| 80 |
+
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape', 'flip',
|
| 81 |
+
'flip_direction'))
|
| 82 |
+
]
|
| 83 |
+
|
| 84 |
+
train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
|
| 85 |
+
|
| 86 |
+
test_pipeline = [
|
| 87 |
+
dict(type='LoadImageFromFile', file_client_args=_base_.file_client_args),
|
| 88 |
+
dict(type='YOLOv5KeepRatioResize', scale=img_scale),
|
| 89 |
+
dict(
|
| 90 |
+
type='LetterResize',
|
| 91 |
+
scale=img_scale,
|
| 92 |
+
allow_scale_up=False,
|
| 93 |
+
pad_val=dict(img=114)),
|
| 94 |
+
dict(type='LoadAnnotations', with_bbox=True, _scope_='mmdet'),
|
| 95 |
+
dict(
|
| 96 |
+
type='mmdet.PackDetInputs',
|
| 97 |
+
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
|
| 98 |
+
'scale_factor', 'pad_param'))
|
| 99 |
+
]
|
| 100 |
+
|
| 101 |
+
val_dataloader = dict(
|
| 102 |
+
dataset=dict(pipeline=test_pipeline, batch_shapes_cfg=batch_shapes_cfg))
|
| 103 |
+
|
| 104 |
+
test_dataloader = val_dataloader
|
| 105 |
+
|
| 106 |
+
# Config for Test Time Augmentation. (TTA)
|
| 107 |
+
_multiscale_resize_transforms = [
|
| 108 |
+
dict(
|
| 109 |
+
type='Compose',
|
| 110 |
+
transforms=[
|
| 111 |
+
dict(type='YOLOv5KeepRatioResize', scale=s),
|
| 112 |
+
dict(
|
| 113 |
+
type='LetterResize',
|
| 114 |
+
scale=s,
|
| 115 |
+
allow_scale_up=False,
|
| 116 |
+
pad_val=dict(img=114))
|
| 117 |
+
]) for s in tta_img_scales
|
| 118 |
+
]
|
| 119 |
+
|
| 120 |
+
tta_pipeline = [
|
| 121 |
+
dict(type='LoadImageFromFile', file_client_args=_base_.file_client_args),
|
| 122 |
+
dict(
|
| 123 |
+
type='TestTimeAug',
|
| 124 |
+
transforms=[
|
| 125 |
+
_multiscale_resize_transforms,
|
| 126 |
+
[
|
| 127 |
+
dict(type='mmdet.RandomFlip', prob=1.),
|
| 128 |
+
dict(type='mmdet.RandomFlip', prob=0.)
|
| 129 |
+
], [dict(type='mmdet.LoadAnnotations', with_bbox=True)],
|
| 130 |
+
[
|
| 131 |
+
dict(
|
| 132 |
+
type='mmdet.PackDetInputs',
|
| 133 |
+
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
|
| 134 |
+
'scale_factor', 'pad_param', 'flip',
|
| 135 |
+
'flip_direction'))
|
| 136 |
+
]
|
| 137 |
+
])
|
| 138 |
+
]
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_s-v61_fast_1xb12-40e_608x352_cat.py
ADDED
|
@@ -0,0 +1,70 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = 'yolov5_s-v61_fast_1xb12-40e_cat.py'
|
| 2 |
+
|
| 3 |
+
# This configuration is used to provide non-square training examples
|
| 4 |
+
# Must be a multiple of 32
|
| 5 |
+
img_scale = (608, 352) # w h
|
| 6 |
+
|
| 7 |
+
anchors = [
|
| 8 |
+
[(65, 35), (159, 45), (119, 80)], # P3/8
|
| 9 |
+
[(215, 77), (224, 116), (170, 166)], # P4/16
|
| 10 |
+
[(376, 108), (339, 176), (483, 190)] # P5/32
|
| 11 |
+
]
|
| 12 |
+
|
| 13 |
+
# ===============================Unmodified in most cases====================
|
| 14 |
+
_base_.model.bbox_head.loss_obj.loss_weight = 1.0 * ((img_scale[1] / 640)**2)
|
| 15 |
+
_base_.model.bbox_head.prior_generator.base_sizes = anchors
|
| 16 |
+
|
| 17 |
+
train_pipeline = [
|
| 18 |
+
*_base_.pre_transform,
|
| 19 |
+
dict(
|
| 20 |
+
type='Mosaic',
|
| 21 |
+
img_scale=img_scale,
|
| 22 |
+
pad_val=114.0,
|
| 23 |
+
pre_transform=_base_.pre_transform),
|
| 24 |
+
dict(
|
| 25 |
+
type='YOLOv5RandomAffine',
|
| 26 |
+
max_rotate_degree=0.0,
|
| 27 |
+
max_shear_degree=0.0,
|
| 28 |
+
scaling_ratio_range=(1 - _base_.affine_scale, 1 + _base_.affine_scale),
|
| 29 |
+
# img_scale is (width, height)
|
| 30 |
+
border=(-img_scale[0] // 2, -img_scale[1] // 2),
|
| 31 |
+
border_val=(114, 114, 114)),
|
| 32 |
+
dict(
|
| 33 |
+
type='mmdet.Albu',
|
| 34 |
+
transforms=_base_.albu_train_transforms,
|
| 35 |
+
bbox_params=dict(
|
| 36 |
+
type='BboxParams',
|
| 37 |
+
format='pascal_voc',
|
| 38 |
+
label_fields=['gt_bboxes_labels', 'gt_ignore_flags']),
|
| 39 |
+
keymap={
|
| 40 |
+
'img': 'image',
|
| 41 |
+
'gt_bboxes': 'bboxes'
|
| 42 |
+
}),
|
| 43 |
+
dict(type='YOLOv5HSVRandomAug'),
|
| 44 |
+
dict(type='mmdet.RandomFlip', prob=0.5),
|
| 45 |
+
dict(
|
| 46 |
+
type='mmdet.PackDetInputs',
|
| 47 |
+
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape', 'flip',
|
| 48 |
+
'flip_direction'))
|
| 49 |
+
]
|
| 50 |
+
|
| 51 |
+
_base_.train_dataloader.dataset.pipeline = train_pipeline
|
| 52 |
+
|
| 53 |
+
test_pipeline = [
|
| 54 |
+
dict(type='LoadImageFromFile', file_client_args=_base_.file_client_args),
|
| 55 |
+
dict(type='YOLOv5KeepRatioResize', scale=img_scale),
|
| 56 |
+
dict(
|
| 57 |
+
type='LetterResize',
|
| 58 |
+
scale=img_scale,
|
| 59 |
+
allow_scale_up=False,
|
| 60 |
+
pad_val=dict(img=114)),
|
| 61 |
+
dict(type='mmdet.LoadAnnotations', with_bbox=True),
|
| 62 |
+
dict(
|
| 63 |
+
type='mmdet.PackDetInputs',
|
| 64 |
+
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
|
| 65 |
+
'scale_factor', 'pad_param'))
|
| 66 |
+
]
|
| 67 |
+
|
| 68 |
+
val_dataloader = dict(
|
| 69 |
+
dataset=dict(pipeline=test_pipeline, batch_shapes_cfg=None))
|
| 70 |
+
test_dataloader = val_dataloader
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_s-v61_fast_1xb12-40e_cat.py
ADDED
|
@@ -0,0 +1,56 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = 'yolov5_s-v61_syncbn_fast_8xb16-300e_coco.py'
|
| 2 |
+
|
| 3 |
+
data_root = './data/cat/'
|
| 4 |
+
class_name = ('cat', )
|
| 5 |
+
num_classes = len(class_name)
|
| 6 |
+
metainfo = dict(classes=class_name, palette=[(20, 220, 60)])
|
| 7 |
+
|
| 8 |
+
anchors = [
|
| 9 |
+
[(68, 69), (154, 91), (143, 162)], # P3/8
|
| 10 |
+
[(242, 160), (189, 287), (391, 207)], # P4/16
|
| 11 |
+
[(353, 337), (539, 341), (443, 432)] # P5/32
|
| 12 |
+
]
|
| 13 |
+
|
| 14 |
+
max_epochs = 40
|
| 15 |
+
train_batch_size_per_gpu = 12
|
| 16 |
+
train_num_workers = 4
|
| 17 |
+
|
| 18 |
+
load_from = 'https://download.openmmlab.com/mmyolo/v0/yolov5/yolov5_s-v61_syncbn_fast_8xb16-300e_coco/yolov5_s-v61_syncbn_fast_8xb16-300e_coco_20220918_084700-86e02187.pth' # noqa
|
| 19 |
+
|
| 20 |
+
model = dict(
|
| 21 |
+
backbone=dict(frozen_stages=4),
|
| 22 |
+
bbox_head=dict(
|
| 23 |
+
head_module=dict(num_classes=num_classes),
|
| 24 |
+
prior_generator=dict(base_sizes=anchors)))
|
| 25 |
+
|
| 26 |
+
train_dataloader = dict(
|
| 27 |
+
batch_size=train_batch_size_per_gpu,
|
| 28 |
+
num_workers=train_num_workers,
|
| 29 |
+
dataset=dict(
|
| 30 |
+
data_root=data_root,
|
| 31 |
+
metainfo=metainfo,
|
| 32 |
+
ann_file='annotations/trainval.json',
|
| 33 |
+
data_prefix=dict(img='images/')))
|
| 34 |
+
|
| 35 |
+
val_dataloader = dict(
|
| 36 |
+
dataset=dict(
|
| 37 |
+
metainfo=metainfo,
|
| 38 |
+
data_root=data_root,
|
| 39 |
+
ann_file='annotations/test.json',
|
| 40 |
+
data_prefix=dict(img='images/')))
|
| 41 |
+
|
| 42 |
+
test_dataloader = val_dataloader
|
| 43 |
+
|
| 44 |
+
_base_.optim_wrapper.optimizer.batch_size_per_gpu = train_batch_size_per_gpu
|
| 45 |
+
|
| 46 |
+
val_evaluator = dict(ann_file=data_root + 'annotations/test.json')
|
| 47 |
+
test_evaluator = val_evaluator
|
| 48 |
+
|
| 49 |
+
default_hooks = dict(
|
| 50 |
+
checkpoint=dict(interval=10, max_keep_ckpts=2, save_best='auto'),
|
| 51 |
+
# The warmup_mim_iter parameter is critical.
|
| 52 |
+
# The default value is 1000 which is not suitable for cat datasets.
|
| 53 |
+
param_scheduler=dict(max_epochs=max_epochs, warmup_mim_iter=10),
|
| 54 |
+
logger=dict(type='LoggerHook', interval=5))
|
| 55 |
+
train_cfg = dict(max_epochs=max_epochs, val_interval=10)
|
| 56 |
+
# visualizer = dict(vis_backends = [dict(type='LocalVisBackend'), dict(type='WandbVisBackend')]) # noqa
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_s-v61_syncbn-detect_8xb16-300e_coco.py
ADDED
|
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = 'yolov5_s-v61_syncbn_8xb16-300e_coco.py'
|
| 2 |
+
|
| 3 |
+
test_pipeline = [
|
| 4 |
+
dict(type='LoadImageFromFile', file_client_args=_base_.file_client_args),
|
| 5 |
+
dict(
|
| 6 |
+
type='LetterResize',
|
| 7 |
+
scale=_base_.img_scale,
|
| 8 |
+
allow_scale_up=True,
|
| 9 |
+
use_mini_pad=True),
|
| 10 |
+
dict(type='LoadAnnotations', with_bbox=True),
|
| 11 |
+
dict(
|
| 12 |
+
type='mmdet.PackDetInputs',
|
| 13 |
+
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
|
| 14 |
+
'scale_factor', 'pad_param'))
|
| 15 |
+
]
|
| 16 |
+
|
| 17 |
+
val_dataloader = dict(
|
| 18 |
+
dataset=dict(pipeline=test_pipeline, batch_shapes_cfg=None))
|
| 19 |
+
test_dataloader = val_dataloader
|
| 20 |
+
|
| 21 |
+
model = dict(
|
| 22 |
+
test_cfg=dict(
|
| 23 |
+
multi_label=False, score_thr=0.25, nms=dict(iou_threshold=0.45)))
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_s-v61_syncbn_8xb16-300e_coco.py
ADDED
|
@@ -0,0 +1,292 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = ['../_base_/default_runtime.py', '../_base_/det_p5_tta.py']
|
| 2 |
+
|
| 3 |
+
# ========================Frequently modified parameters======================
|
| 4 |
+
# -----data related-----
|
| 5 |
+
data_root = 'data/coco/' # Root path of data
|
| 6 |
+
# Path of train annotation file
|
| 7 |
+
train_ann_file = 'annotations/instances_train2017.json'
|
| 8 |
+
train_data_prefix = 'train2017/' # Prefix of train image path
|
| 9 |
+
# Path of val annotation file
|
| 10 |
+
val_ann_file = 'annotations/instances_val2017.json'
|
| 11 |
+
val_data_prefix = 'val2017/' # Prefix of val image path
|
| 12 |
+
|
| 13 |
+
num_classes = 80 # Number of classes for classification
|
| 14 |
+
# Batch size of a single GPU during training
|
| 15 |
+
train_batch_size_per_gpu = 16
|
| 16 |
+
# Worker to pre-fetch data for each single GPU during training
|
| 17 |
+
train_num_workers = 8
|
| 18 |
+
# persistent_workers must be False if num_workers is 0
|
| 19 |
+
persistent_workers = True
|
| 20 |
+
|
| 21 |
+
# -----model related-----
|
| 22 |
+
# Basic size of multi-scale prior box
|
| 23 |
+
anchors = [
|
| 24 |
+
[(10, 13), (16, 30), (33, 23)], # P3/8
|
| 25 |
+
[(30, 61), (62, 45), (59, 119)], # P4/16
|
| 26 |
+
[(116, 90), (156, 198), (373, 326)] # P5/32
|
| 27 |
+
]
|
| 28 |
+
|
| 29 |
+
# -----train val related-----
|
| 30 |
+
# Base learning rate for optim_wrapper. Corresponding to 8xb16=128 bs
|
| 31 |
+
base_lr = 0.01
|
| 32 |
+
max_epochs = 300 # Maximum training epochs
|
| 33 |
+
|
| 34 |
+
model_test_cfg = dict(
|
| 35 |
+
# The config of multi-label for multi-class prediction.
|
| 36 |
+
multi_label=True,
|
| 37 |
+
# The number of boxes before NMS
|
| 38 |
+
nms_pre=30000,
|
| 39 |
+
score_thr=0.001, # Threshold to filter out boxes.
|
| 40 |
+
nms=dict(type='nms', iou_threshold=0.65), # NMS type and threshold
|
| 41 |
+
max_per_img=300) # Max number of detections of each image
|
| 42 |
+
|
| 43 |
+
# ========================Possible modified parameters========================
|
| 44 |
+
# -----data related-----
|
| 45 |
+
img_scale = (640, 640) # width, height
|
| 46 |
+
# Dataset type, this will be used to define the dataset
|
| 47 |
+
dataset_type = 'YOLOv5CocoDataset'
|
| 48 |
+
# Batch size of a single GPU during validation
|
| 49 |
+
val_batch_size_per_gpu = 1
|
| 50 |
+
# Worker to pre-fetch data for each single GPU during validation
|
| 51 |
+
val_num_workers = 2
|
| 52 |
+
|
| 53 |
+
# Config of batch shapes. Only on val.
|
| 54 |
+
# It means not used if batch_shapes_cfg is None.
|
| 55 |
+
batch_shapes_cfg = dict(
|
| 56 |
+
type='BatchShapePolicy',
|
| 57 |
+
batch_size=val_batch_size_per_gpu,
|
| 58 |
+
img_size=img_scale[0],
|
| 59 |
+
# The image scale of padding should be divided by pad_size_divisor
|
| 60 |
+
size_divisor=32,
|
| 61 |
+
# Additional paddings for pixel scale
|
| 62 |
+
extra_pad_ratio=0.5)
|
| 63 |
+
|
| 64 |
+
# -----model related-----
|
| 65 |
+
# The scaling factor that controls the depth of the network structure
|
| 66 |
+
deepen_factor = 0.33
|
| 67 |
+
# The scaling factor that controls the width of the network structure
|
| 68 |
+
widen_factor = 0.5
|
| 69 |
+
# Strides of multi-scale prior box
|
| 70 |
+
strides = [8, 16, 32]
|
| 71 |
+
num_det_layers = 3 # The number of model output scales
|
| 72 |
+
norm_cfg = dict(type='BN', momentum=0.03, eps=0.001) # Normalization config
|
| 73 |
+
|
| 74 |
+
# -----train val related-----
|
| 75 |
+
affine_scale = 0.5 # YOLOv5RandomAffine scaling ratio
|
| 76 |
+
loss_cls_weight = 0.5
|
| 77 |
+
loss_bbox_weight = 0.05
|
| 78 |
+
loss_obj_weight = 1.0
|
| 79 |
+
prior_match_thr = 4. # Priori box matching threshold
|
| 80 |
+
# The obj loss weights of the three output layers
|
| 81 |
+
obj_level_weights = [4., 1., 0.4]
|
| 82 |
+
lr_factor = 0.01 # Learning rate scaling factor
|
| 83 |
+
weight_decay = 0.0005
|
| 84 |
+
# Save model checkpoint and validation intervals
|
| 85 |
+
save_checkpoint_intervals = 10
|
| 86 |
+
# The maximum checkpoints to keep.
|
| 87 |
+
max_keep_ckpts = 3
|
| 88 |
+
# Single-scale training is recommended to
|
| 89 |
+
# be turned on, which can speed up training.
|
| 90 |
+
env_cfg = dict(cudnn_benchmark=True)
|
| 91 |
+
|
| 92 |
+
# ===============================Unmodified in most cases====================
|
| 93 |
+
model = dict(
|
| 94 |
+
type='YOLODetector',
|
| 95 |
+
data_preprocessor=dict(
|
| 96 |
+
type='mmdet.DetDataPreprocessor',
|
| 97 |
+
mean=[0., 0., 0.],
|
| 98 |
+
std=[255., 255., 255.],
|
| 99 |
+
bgr_to_rgb=True),
|
| 100 |
+
backbone=dict(
|
| 101 |
+
type='YOLOv5CSPDarknet',
|
| 102 |
+
deepen_factor=deepen_factor,
|
| 103 |
+
widen_factor=widen_factor,
|
| 104 |
+
norm_cfg=norm_cfg,
|
| 105 |
+
act_cfg=dict(type='SiLU', inplace=True)),
|
| 106 |
+
neck=dict(
|
| 107 |
+
type='YOLOv5PAFPN',
|
| 108 |
+
deepen_factor=deepen_factor,
|
| 109 |
+
widen_factor=widen_factor,
|
| 110 |
+
in_channels=[256, 512, 1024],
|
| 111 |
+
out_channels=[256, 512, 1024],
|
| 112 |
+
num_csp_blocks=3,
|
| 113 |
+
norm_cfg=norm_cfg,
|
| 114 |
+
act_cfg=dict(type='SiLU', inplace=True)),
|
| 115 |
+
bbox_head=dict(
|
| 116 |
+
type='YOLOv5Head',
|
| 117 |
+
head_module=dict(
|
| 118 |
+
type='YOLOv5HeadModule',
|
| 119 |
+
num_classes=num_classes,
|
| 120 |
+
in_channels=[256, 512, 1024],
|
| 121 |
+
widen_factor=widen_factor,
|
| 122 |
+
featmap_strides=strides,
|
| 123 |
+
num_base_priors=3),
|
| 124 |
+
prior_generator=dict(
|
| 125 |
+
type='mmdet.YOLOAnchorGenerator',
|
| 126 |
+
base_sizes=anchors,
|
| 127 |
+
strides=strides),
|
| 128 |
+
# scaled based on number of detection layers
|
| 129 |
+
loss_cls=dict(
|
| 130 |
+
type='mmdet.CrossEntropyLoss',
|
| 131 |
+
use_sigmoid=True,
|
| 132 |
+
reduction='mean',
|
| 133 |
+
loss_weight=loss_cls_weight *
|
| 134 |
+
(num_classes / 80 * 3 / num_det_layers)),
|
| 135 |
+
loss_bbox=dict(
|
| 136 |
+
type='IoULoss',
|
| 137 |
+
iou_mode='ciou',
|
| 138 |
+
bbox_format='xywh',
|
| 139 |
+
eps=1e-7,
|
| 140 |
+
reduction='mean',
|
| 141 |
+
loss_weight=loss_bbox_weight * (3 / num_det_layers),
|
| 142 |
+
return_iou=True),
|
| 143 |
+
loss_obj=dict(
|
| 144 |
+
type='mmdet.CrossEntropyLoss',
|
| 145 |
+
use_sigmoid=True,
|
| 146 |
+
reduction='mean',
|
| 147 |
+
loss_weight=loss_obj_weight *
|
| 148 |
+
((img_scale[0] / 640)**2 * 3 / num_det_layers)),
|
| 149 |
+
prior_match_thr=prior_match_thr,
|
| 150 |
+
obj_level_weights=obj_level_weights),
|
| 151 |
+
test_cfg=model_test_cfg)
|
| 152 |
+
|
| 153 |
+
albu_train_transforms = [
|
| 154 |
+
dict(type='Blur', p=0.01),
|
| 155 |
+
dict(type='MedianBlur', p=0.01),
|
| 156 |
+
dict(type='ToGray', p=0.01),
|
| 157 |
+
dict(type='CLAHE', p=0.01)
|
| 158 |
+
]
|
| 159 |
+
|
| 160 |
+
pre_transform = [
|
| 161 |
+
dict(type='LoadImageFromFile', file_client_args=_base_.file_client_args),
|
| 162 |
+
dict(type='LoadAnnotations', with_bbox=True)
|
| 163 |
+
]
|
| 164 |
+
|
| 165 |
+
train_pipeline = [
|
| 166 |
+
*pre_transform,
|
| 167 |
+
dict(
|
| 168 |
+
type='Mosaic',
|
| 169 |
+
img_scale=img_scale,
|
| 170 |
+
pad_val=114.0,
|
| 171 |
+
pre_transform=pre_transform),
|
| 172 |
+
dict(
|
| 173 |
+
type='YOLOv5RandomAffine',
|
| 174 |
+
max_rotate_degree=0.0,
|
| 175 |
+
max_shear_degree=0.0,
|
| 176 |
+
scaling_ratio_range=(1 - affine_scale, 1 + affine_scale),
|
| 177 |
+
# img_scale is (width, height)
|
| 178 |
+
border=(-img_scale[0] // 2, -img_scale[1] // 2),
|
| 179 |
+
border_val=(114, 114, 114)),
|
| 180 |
+
dict(
|
| 181 |
+
type='mmdet.Albu',
|
| 182 |
+
transforms=albu_train_transforms,
|
| 183 |
+
bbox_params=dict(
|
| 184 |
+
type='BboxParams',
|
| 185 |
+
format='pascal_voc',
|
| 186 |
+
label_fields=['gt_bboxes_labels', 'gt_ignore_flags']),
|
| 187 |
+
keymap={
|
| 188 |
+
'img': 'image',
|
| 189 |
+
'gt_bboxes': 'bboxes'
|
| 190 |
+
}),
|
| 191 |
+
dict(type='YOLOv5HSVRandomAug'),
|
| 192 |
+
dict(type='mmdet.RandomFlip', prob=0.5),
|
| 193 |
+
dict(
|
| 194 |
+
type='mmdet.PackDetInputs',
|
| 195 |
+
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape', 'flip',
|
| 196 |
+
'flip_direction'))
|
| 197 |
+
]
|
| 198 |
+
|
| 199 |
+
train_dataloader = dict(
|
| 200 |
+
batch_size=train_batch_size_per_gpu,
|
| 201 |
+
num_workers=train_num_workers,
|
| 202 |
+
persistent_workers=persistent_workers,
|
| 203 |
+
pin_memory=True,
|
| 204 |
+
sampler=dict(type='DefaultSampler', shuffle=True),
|
| 205 |
+
dataset=dict(
|
| 206 |
+
type=dataset_type,
|
| 207 |
+
data_root=data_root,
|
| 208 |
+
ann_file=train_ann_file,
|
| 209 |
+
data_prefix=dict(img=train_data_prefix),
|
| 210 |
+
filter_cfg=dict(filter_empty_gt=False, min_size=32),
|
| 211 |
+
pipeline=train_pipeline))
|
| 212 |
+
|
| 213 |
+
test_pipeline = [
|
| 214 |
+
dict(type='LoadImageFromFile', file_client_args=_base_.file_client_args),
|
| 215 |
+
dict(type='YOLOv5KeepRatioResize', scale=img_scale),
|
| 216 |
+
dict(
|
| 217 |
+
type='LetterResize',
|
| 218 |
+
scale=img_scale,
|
| 219 |
+
allow_scale_up=False,
|
| 220 |
+
pad_val=dict(img=114)),
|
| 221 |
+
dict(type='LoadAnnotations', with_bbox=True, _scope_='mmdet'),
|
| 222 |
+
dict(
|
| 223 |
+
type='mmdet.PackDetInputs',
|
| 224 |
+
meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
|
| 225 |
+
'scale_factor', 'pad_param'))
|
| 226 |
+
]
|
| 227 |
+
|
| 228 |
+
val_dataloader = dict(
|
| 229 |
+
batch_size=val_batch_size_per_gpu,
|
| 230 |
+
num_workers=val_num_workers,
|
| 231 |
+
persistent_workers=persistent_workers,
|
| 232 |
+
pin_memory=True,
|
| 233 |
+
drop_last=False,
|
| 234 |
+
sampler=dict(type='DefaultSampler', shuffle=False),
|
| 235 |
+
dataset=dict(
|
| 236 |
+
type=dataset_type,
|
| 237 |
+
data_root=data_root,
|
| 238 |
+
test_mode=True,
|
| 239 |
+
data_prefix=dict(img=val_data_prefix),
|
| 240 |
+
ann_file=val_ann_file,
|
| 241 |
+
pipeline=test_pipeline,
|
| 242 |
+
batch_shapes_cfg=batch_shapes_cfg))
|
| 243 |
+
|
| 244 |
+
test_dataloader = val_dataloader
|
| 245 |
+
|
| 246 |
+
param_scheduler = None
|
| 247 |
+
optim_wrapper = dict(
|
| 248 |
+
type='OptimWrapper',
|
| 249 |
+
optimizer=dict(
|
| 250 |
+
type='SGD',
|
| 251 |
+
lr=base_lr,
|
| 252 |
+
momentum=0.937,
|
| 253 |
+
weight_decay=weight_decay,
|
| 254 |
+
nesterov=True,
|
| 255 |
+
batch_size_per_gpu=train_batch_size_per_gpu),
|
| 256 |
+
constructor='YOLOv5OptimizerConstructor')
|
| 257 |
+
|
| 258 |
+
default_hooks = dict(
|
| 259 |
+
param_scheduler=dict(
|
| 260 |
+
type='YOLOv5ParamSchedulerHook',
|
| 261 |
+
scheduler_type='linear',
|
| 262 |
+
lr_factor=lr_factor,
|
| 263 |
+
max_epochs=max_epochs),
|
| 264 |
+
checkpoint=dict(
|
| 265 |
+
type='CheckpointHook',
|
| 266 |
+
interval=save_checkpoint_intervals,
|
| 267 |
+
save_best='auto',
|
| 268 |
+
max_keep_ckpts=max_keep_ckpts))
|
| 269 |
+
|
| 270 |
+
custom_hooks = [
|
| 271 |
+
dict(
|
| 272 |
+
type='EMAHook',
|
| 273 |
+
ema_type='ExpMomentumEMA',
|
| 274 |
+
momentum=0.0001,
|
| 275 |
+
update_buffers=True,
|
| 276 |
+
strict_load=False,
|
| 277 |
+
priority=49)
|
| 278 |
+
]
|
| 279 |
+
|
| 280 |
+
val_evaluator = dict(
|
| 281 |
+
type='mmdet.CocoMetric',
|
| 282 |
+
proposal_nums=(100, 1, 10),
|
| 283 |
+
ann_file=data_root + val_ann_file,
|
| 284 |
+
metric='bbox')
|
| 285 |
+
test_evaluator = val_evaluator
|
| 286 |
+
|
| 287 |
+
train_cfg = dict(
|
| 288 |
+
type='EpochBasedTrainLoop',
|
| 289 |
+
max_epochs=max_epochs,
|
| 290 |
+
val_interval=save_checkpoint_intervals)
|
| 291 |
+
val_cfg = dict(type='ValLoop')
|
| 292 |
+
test_cfg = dict(type='TestLoop')
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_s-v61_syncbn_fast_1xb4-300e_balloon.py
ADDED
|
@@ -0,0 +1,42 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = './yolov5_s-v61_syncbn_fast_8xb16-300e_coco.py'
|
| 2 |
+
|
| 3 |
+
# ========================modified parameters======================
|
| 4 |
+
data_root = 'data/balloon/'
|
| 5 |
+
# Path of train annotation file
|
| 6 |
+
train_ann_file = 'train.json'
|
| 7 |
+
train_data_prefix = 'train/' # Prefix of train image path
|
| 8 |
+
# Path of val annotation file
|
| 9 |
+
val_ann_file = 'val.json'
|
| 10 |
+
val_data_prefix = 'val/' # Prefix of val image path
|
| 11 |
+
metainfo = {
|
| 12 |
+
'classes': ('balloon', ),
|
| 13 |
+
'palette': [
|
| 14 |
+
(220, 20, 60),
|
| 15 |
+
]
|
| 16 |
+
}
|
| 17 |
+
num_classes = 1
|
| 18 |
+
|
| 19 |
+
train_batch_size_per_gpu = 4
|
| 20 |
+
train_num_workers = 2
|
| 21 |
+
log_interval = 1
|
| 22 |
+
|
| 23 |
+
# =======================Unmodified in most cases==================
|
| 24 |
+
train_dataloader = dict(
|
| 25 |
+
batch_size=train_batch_size_per_gpu,
|
| 26 |
+
num_workers=train_num_workers,
|
| 27 |
+
dataset=dict(
|
| 28 |
+
data_root=data_root,
|
| 29 |
+
metainfo=metainfo,
|
| 30 |
+
data_prefix=dict(img=train_data_prefix),
|
| 31 |
+
ann_file=train_ann_file))
|
| 32 |
+
val_dataloader = dict(
|
| 33 |
+
dataset=dict(
|
| 34 |
+
data_root=data_root,
|
| 35 |
+
metainfo=metainfo,
|
| 36 |
+
data_prefix=dict(img=val_data_prefix),
|
| 37 |
+
ann_file=val_ann_file))
|
| 38 |
+
test_dataloader = val_dataloader
|
| 39 |
+
val_evaluator = dict(ann_file=data_root + val_ann_file)
|
| 40 |
+
test_evaluator = val_evaluator
|
| 41 |
+
model = dict(bbox_head=dict(head_module=dict(num_classes=num_classes)))
|
| 42 |
+
default_hooks = dict(logger=dict(interval=log_interval))
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_s-v61_syncbn_fast_8xb16-300e_coco.py
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = 'yolov5_s-v61_syncbn_8xb16-300e_coco.py'
|
| 2 |
+
|
| 3 |
+
# fast means faster training speed,
|
| 4 |
+
# but less flexibility for multitasking
|
| 5 |
+
model = dict(
|
| 6 |
+
data_preprocessor=dict(
|
| 7 |
+
type='YOLOv5DetDataPreprocessor',
|
| 8 |
+
mean=[0., 0., 0.],
|
| 9 |
+
std=[255., 255., 255.],
|
| 10 |
+
bgr_to_rgb=True))
|
| 11 |
+
|
| 12 |
+
train_dataloader = dict(collate_fn=dict(type='yolov5_collate'))
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_x-p6-v62_syncbn_fast_8xb16-300e_coco.py
ADDED
|
@@ -0,0 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = './yolov5_m-p6-v62_syncbn_fast_8xb16-300e_coco.py'
|
| 2 |
+
deepen_factor = 1.33
|
| 3 |
+
widen_factor = 1.25
|
| 4 |
+
|
| 5 |
+
model = dict(
|
| 6 |
+
backbone=dict(
|
| 7 |
+
deepen_factor=deepen_factor,
|
| 8 |
+
widen_factor=widen_factor,
|
| 9 |
+
),
|
| 10 |
+
neck=dict(
|
| 11 |
+
deepen_factor=deepen_factor,
|
| 12 |
+
widen_factor=widen_factor,
|
| 13 |
+
),
|
| 14 |
+
bbox_head=dict(head_module=dict(widen_factor=widen_factor)))
|
Model/Model6/Model6_0_ClothesDetection/mmyolo/configs/yolov5/yolov5_x-v61_syncbn_fast_8xb16-300e_coco.py
ADDED
|
@@ -0,0 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
_base_ = './yolov5_m-v61_syncbn_fast_8xb16-300e_coco.py'
|
| 2 |
+
deepen_factor = 1.33
|
| 3 |
+
widen_factor = 1.25
|
| 4 |
+
|
| 5 |
+
model = dict(
|
| 6 |
+
backbone=dict(
|
| 7 |
+
deepen_factor=deepen_factor,
|
| 8 |
+
widen_factor=widen_factor,
|
| 9 |
+
),
|
| 10 |
+
neck=dict(
|
| 11 |
+
deepen_factor=deepen_factor,
|
| 12 |
+
widen_factor=widen_factor,
|
| 13 |
+
),
|
| 14 |
+
bbox_head=dict(head_module=dict(widen_factor=widen_factor)))
|