Update docs (#653)

* Update docs
* Update README.md

README.md

@@ -44,7 +44,14 @@
 | BlazeFace          | ✗      |                             ✗ | ✗          | ✗     | ✗         |  ✗     |  ✗      |
 | Faceboxes          | ✗      |                             ✗ | ✗          | ✗     | ✗         |  ✗     |  ✗      |
 
-<a name="vd">[1]</a> [ResNet-vd](https://arxiv.org/pdf/1812.01187) 模型提供了较大的精度提高和较少的性能损失。
+<a name="vd">[1]</a> [ResNet-vd](https://arxiv.org/pdf/1812.01187) 模型预测速度基本不变的情况下提高了精度。
+
+更多的模型:
+
+- EfficientDet
+- FCOS
+- CornerNet-Squeeze
+- YOLOv4
 
 更多的Backone：
 
@@ -52,10 +59,11 @@
 - VGG
 - GCNet
 - CBNet
+- Hourglass
 
 扩展特性：
 
-- [x] **Synchronized Batch Norm**: 目前在YOLOv3中使用。
+- [x] **Synchronized Batch Norm**
 - [x] **Group Norm**
 - [x] **Modulated Deformable Convolution**
 - [x] **Deformable PSRoI Pooling**
@@ -107,12 +115,14 @@
 ## 模型库
 
 - [模型库](docs/MODEL_ZOO_cn.md)
-- [人脸检测模型](docs/featured_model/FACE_DETECTION.md) 开源BlazeFace系列模型，Wider-Face数据集上最高精度达到91.5%，同时具备了较高的预测性能
-- [行人检测和车辆检测预训练模型](docs/featured_model/CONTRIB_cn.md) 针对不同场景的检测模型
-- [YOLOv3增强模型](docs/featured_model/YOLOv3_ENHANCEMENT.md) 改进原始YOLOv3，精度达到43.6%，原论文精度为33.0%，同时预测速度也得到提升
-- [Objects365 2019 Challenge夺冠模型](docs/featured_model/CACascadeRCNN.md) Objects365 Full Track任务中最好的单模型之一,精度达到31.7%
-- [Open Images V5和Objects365数据集模型](docs/featured_model/OIDV5_BASELINE_MODEL.md)
 - [移动端模型](configs/mobile/README.md)
+- [Anchor free模型](configs/anchor_free/README.md)
+- [人脸检测模型](docs/featured_model/FACE_DETECTION.md)
+- [YOLOv3增强模型](docs/featured_model/YOLOv3_ENHANCEMENT.md): COCO mAP高达43.6%，原论文精度为33.0%
+- [行人检测预训练模型](docs/featured_model/CONTRIB_cn.md)
+- [车辆检测预训练模型](docs/featured_model/CONTRIB_cn.md)
+- [Objects365 2019 Challenge夺冠模型](docs/featured_model/CACascadeRCNN.md)
+- [Open Images 2019-Object Detction比赛最佳单模型](docs/featured_model/OIDV5_BASELINE_MODEL.md)
 
 
 ## 许可证书

README_en.md

@@ -59,6 +59,13 @@ Supported Architectures:
 
 <a name="vd">[1]</a> [ResNet-vd](https://arxiv.org/pdf/1812.01187) models offer much improved accuracy with negligible performance cost.
 
+More models:
+
+- EfficientDet
+- FCOS
+- CornerNet-Squeeze
+- YOLOv4
+
 More Backbones:
 
 - DarkNet
@@ -68,7 +75,7 @@ More Backbones:
 
 Advanced Features:
 
-- [x] **Synchronized Batch Norm**: currently used by YOLOv3.
+- [x] **Synchronized Batch Norm**
 - [x] **Group Norm**
 - [x] **Modulated Deformable Convolution**
 - [x] **Deformable PSRoI Pooling**
@@ -121,12 +128,14 @@ The following is the relationship between COCO mAP and FPS on Tesla V100 of repr
 ## Model Zoo
 
 - Pretrained models are available in the [PaddleDetection model zoo](docs/MODEL_ZOO.md).
-- [Face detection models](docs/featured_model/FACE_DETECTION_en.md) BlazeFace series model with the highest precision of 91.5% on Wider-Face dataset and outstanding inference performance.
-- [Pretrained models for pedestrian and vehicle detection](docs/featured_model/CONTRIB.md) Models for object detection in specific scenarios.
-- [YOLOv3 enhanced model](docs/featured_model/YOLOv3_ENHANCEMENT.md) Compared to MAP of 33.0% in paper, enhanced YOLOv3 reaches the MAP of 43.6% and inference speed is improved as well
-- [Objects365 2019 Challenge champion model](docs/featured_model/CACascadeRCNN.md) One of the best single models in Objects365 Full Track of which MAP reaches 31.7%.
-- [Open Images Dataset V5 and Objects365 Dataset models](docs/featured_model/OIDV5_BASELINE_MODEL.md)
 - [Mobile models](configs/mobile/README.md)
+- [Anchor free models](configs/anchor_free/README.md)
+- [Face detection models](docs/featured_model/FACE_DETECTION_en.md)
+- [Pretrained models for pedestrian detection](docs/featured_model/CONTRIB.md)
+- [Pretrained models for vehicle detection](docs/featured_model/CONTRIB.md)
+- [YOLOv3 enhanced model](docs/featured_model/YOLOv3_ENHANCEMENT.md): Compared to MAP of 33.0% in paper, enhanced YOLOv3 reaches the MAP of 43.6%, and inference speed is improved as well
+- [Objects365 2019 Challenge champion model](docs/featured_model/CACascadeRCNN.md)
+- [Best single model of Open Images 2019-Object Detction](docs/featured_model/OIDV5_BASELINE_MODEL.md)
 
 
 ## License

docs/advanced_tutorials/TRANSFER_LEARNING_cn.md

@@ -2,11 +2,10 @@
 
 迁移学习为利用已有知识，对新知识进行学习。例如利用ImageNet分类预训练模型做初始化来训练检测模型，利用在COCO数据集上的检测模型做初始化来训练基于PascalVOC数据集的检测模型。
 
-在进行迁移学习时，由于会使用不同的数据集，数据类别数与COCO/VOC数据类别不同，导致在加载PaddlePaddle开源模型时，与类别数相关的权重（例如分类模块的fc层）会出现维度不匹配的问题；另外，如果需要结构更加复杂的模型，需要对已有开源模型结构进行调整，对应权重也需要选择性加载。因此，需要检测库能够指定参数字段，在加载模型时不加载匹配的权重。
+在进行迁移学习时，由于会使用不同的数据集，数据类别数与COCO/VOC数据类别不同，导致在加载开源模型(如COCO预训练模型)时，与类别数相关的权重（例如分类模块的fc层）会出现维度不匹配的问题；另外，如果需要结构更加复杂的模型，需要对已有开源模型结构进行调整，对应权重也需要选择性加载。因此，需要检测库能够指定参数字段，在加载模型时不加载匹配的权重。
 
-## PaddleDetection进行迁移学习
 
-在迁移学习中，对预训练模型进行选择性加载，PaddleDetection支持如下两种迁移学习方式：
+在迁移学习中，对预训练模型进行选择性加载，支持如下两种迁移学习方式：
 
 #### 直接加载预训练权重（**推荐方式**）
 
@@ -38,7 +37,7 @@ python -u tools/train.py -c configs/faster_rcnn_r50_1x.yml \
 1. pretrain\_weights的路径为COCO数据集上开源的faster RCNN模型链接，完整模型链接可参考[MODEL_ZOO](../MODEL_ZOO_cn.md)
 2. finetune\_exclude\_pretrained\_params中设置参数字段，如果参数名能够匹配以上参数字段（通配符匹配方式），则在模型加载时忽略该参数。
 
-如果用户需要利用自己的数据进行finetune，模型结构不变，只需要忽略与类别数相关的参数。PaddleDetection给出了不同模型类型所对应的忽略参数字段。如下表所示：</br>
+如果用户需要利用自己的数据进行finetune，模型结构不变，只需要忽略与类别数相关的参数，不同模型类型所对应的忽略参数字段如下表所示：</br>
 
 |      模型类型      |             忽略参数字段                  |
 | :----------------: | :---------------------------------------: |

docs/tutorials/GETTING_STARTED.md

@@ -11,6 +11,8 @@ instructions](INSTALL.md).
 PaddleDetection provides scripots for training, evalution and inference with various features according to different configure.
 
 ```bash
+# set PYTHONPATH
+export PYTHONPATH=$PYTHONPATH:.
 # training in single-GPU and multi-GPU. specify different GPU numbers by CUDA_VISIBLE_DEVICES
 export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
 python tools/train.py -c configs/faster_rcnn_r50_1x.yml

docs/tutorials/GETTING_STARTED_cn.md

@@ -8,6 +8,8 @@
 PaddleDetection提供了训练/评估/推断三个功能的使用脚本，支持通过不同可选参数实现特定功能
 
 ```bash
+# 设置PYTHONPATH路径
+export PYTHONPATH=$PYTHONPATH:.
 # GPU训练 支持单卡，多卡训练，通过CUDA_VISIBLE_DEVICES指定卡号
 export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
 python tools/train.py -c configs/faster_rcnn_r50_1x.yml

slim/distillation/distill.py

@@ -16,11 +16,16 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-import os
+import os, sys
+# add python path of PadleDetection to sys.path
+parent_path = os.path.abspath(os.path.join(__file__, *(['..'] * 3)))
+if parent_path not in sys.path:
+    sys.path.append(parent_path)
+
 import numpy as np
 from collections import OrderedDict
-from paddleslim.dist.single_distiller import merge, l2_loss
 
+from paddleslim.dist.single_distiller import merge, l2_loss
 from paddle import fluid
 from ppdet.core.workspace import load_config, merge_config, create
 from ppdet.data.reader import create_reader

slim/extensions/distill_pruned_model/distill_pruned_model.py

@@ -16,7 +16,13 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-import os
+import os, sys
+
+# add python path of PadleDetection to sys.path
+parent_path = os.path.abspath(os.path.join(__file__, *(['..'] * 4)))
+if parent_path not in sys.path:
+    sys.path.append(parent_path)
+
 import numpy as np
 from collections import OrderedDict
 from paddleslim.dist.single_distiller import merge, l2_loss

slim/nas/train_nas.py

@@ -16,29 +16,19 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-import os
+import os, sys
+# add python path of PadleDetection to sys.path
+parent_path = os.path.abspath(os.path.join(__file__, *(['..'] * 3)))
+if parent_path not in sys.path:
+    sys.path.append(parent_path)
+
 import time
 import numpy as np
 import datetime
 from collections import deque
 
-
-def set_paddle_flags(**kwargs):
-    for key, value in kwargs.items():
-        if os.environ.get(key, None) is None:
-            os.environ[key] = str(value)
-
-
-# NOTE(paddle-dev): All of these flags should be set before
-# `import paddle`. Otherwise, it would not take any effect.
-set_paddle_flags(
-    FLAGS_eager_delete_tensor_gb=0,  # enable GC to save memory
-)
-
 from paddle import fluid
-import sys
 
-sys.path.append("../../")
 from ppdet.experimental import mixed_precision_context
 from ppdet.core.workspace import load_config, merge_config, create, register
 from ppdet.data.reader import create_reader

slim/prune/eval.py

@@ -16,20 +16,12 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-import os
+import os, sys
 
-
-def set_paddle_flags(**kwargs):
-    for key, value in kwargs.items():
-        if os.environ.get(key, None) is None:
-            os.environ[key] = str(value)
-
-
-# NOTE(paddle-dev): All of these flags should be set before
-# `import paddle`. Otherwise, it would not take any effect.
-set_paddle_flags(
-    FLAGS_eager_delete_tensor_gb=0,  # enable GC to save memory
-)
+# add python path of PadleDetection to sys.path
+parent_path = os.path.abspath(os.path.join(__file__, *(['..'] * 3)))
+if parent_path not in sys.path:
+    sys.path.append(parent_path)
 
 import paddle.fluid as fluid
 from paddleslim.prune import Pruner

slim/prune/export_model.py

@@ -16,7 +16,11 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-import os
+import os, sys
+# add python path of PadleDetection to sys.path
+parent_path = os.path.abspath(os.path.join(__file__, *(['..'] * 3)))
+if parent_path not in sys.path:
+    sys.path.append(parent_path)
 
 from paddle import fluid
 

slim/prune/prune.py

@@ -16,7 +16,13 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-import os
+import os, sys
+
+# add python path of PadleDetection to sys.path
+parent_path = os.path.abspath(os.path.join(__file__, *(['..'] * 3)))
+if parent_path not in sys.path:
+    sys.path.append(parent_path)
+
 import time
 import numpy as np
 import datetime
@@ -24,6 +30,7 @@ from collections import deque
 from paddleslim.prune import Pruner
 from paddleslim.analysis import flops
 from paddle import fluid
+
 from ppdet.experimental import mixed_precision_context
 from ppdet.core.workspace import load_config, merge_config, create
 from ppdet.data.reader import create_reader

slim/quantization/eval.py

@@ -16,8 +16,12 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-import os
-import sys
+import os, sys
+
+# add python path of PadleDetection to sys.path
+parent_path = os.path.abspath(os.path.join(__file__, *(['..'] * 3)))
+if parent_path not in sys.path:
+    sys.path.append(parent_path)
 
 import paddle.fluid as fluid
 

slim/quantization/export_model.py

@@ -16,8 +16,12 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-import os
-import sys
+import os, sys
+# add python path of PadleDetection to sys.path
+parent_path = os.path.abspath(os.path.join(__file__, *(['..'] * 3)))
+if parent_path not in sys.path:
+    sys.path.append(parent_path)
+
 from paddle import fluid
 
 from ppdet.core.workspace import load_config, merge_config, create

slim/quantization/infer.py

@@ -16,17 +16,19 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-import os
-import glob
-import sys
+import os, sys
+# add python path of PadleDetection to sys.path
+parent_path = os.path.abspath(os.path.join(__file__, *(['..'] * 3)))
+if parent_path not in sys.path:
+    sys.path.append(parent_path)
 
+import glob
 import numpy as np
 from PIL import Image
 
 from paddle import fluid
 
 from ppdet.core.workspace import load_config, merge_config, create
-
 from ppdet.utils.eval_utils import parse_fetches
 from ppdet.utils.cli import ArgsParser
 from ppdet.utils.check import check_gpu, check_version, check_config

slim/quantization/train.py

@@ -16,8 +16,12 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-import os
-import sys
+import os, sys
+# add python path of PadleDetection to sys.path
+parent_path = os.path.abspath(os.path.join(__file__, *(['..'] * 3)))
+if parent_path not in sys.path:
+    sys.path.append(parent_path)
+
 import time
 import numpy as np
 import datetime
@@ -28,7 +32,6 @@ from paddle import fluid
 
 from ppdet.core.workspace import load_config, merge_config, create
 from ppdet.data.reader import create_reader
-
 from ppdet.utils import dist_utils
 from ppdet.utils.eval_utils import parse_fetches, eval_run, eval_results
 from ppdet.utils.stats import TrainingStats

slim/sensitive/sensitive.py

@@ -16,25 +16,17 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-import os
+import os, sys
+# add python path of PadleDetection to sys.path
+parent_path = os.path.abspath(os.path.join(__file__, *(['..'] * 3)))
+if parent_path not in sys.path:
+    sys.path.append(parent_path)
+
 import time
 import numpy as np
 import datetime
 from collections import deque
 
-
-def set_paddle_flags(**kwargs):
-    for key, value in kwargs.items():
-        if os.environ.get(key, None) is None:
-            os.environ[key] = str(value)
-
-
-# NOTE(paddle-dev): All of these flags should be set before
-# `import paddle`. Otherwise, it would not take any effect.
-set_paddle_flags(
-    FLAGS_eager_delete_tensor_gb=0,  # enable GC to save memory
-)
-
 from paddle import fluid
 from ppdet.experimental import mixed_precision_context
 from ppdet.core.workspace import load_config, merge_config, create

tools/eval.py

@@ -16,26 +16,12 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-import os
-import sys
+import os, sys
 # add python path of PadleDetection to sys.path
 parent_path = os.path.abspath(os.path.join(__file__, *(['..'] * 2)))
 if parent_path not in sys.path:
     sys.path.append(parent_path)
 
-
-def set_paddle_flags(**kwargs):
-    for key, value in kwargs.items():
-        if os.environ.get(key, None) is None:
-            os.environ[key] = str(value)
-
-
-# NOTE(paddle-dev): All of these flags should be set before
-# `import paddle`. Otherwise, it would not take any effect.
-set_paddle_flags(
-    FLAGS_eager_delete_tensor_gb=0,  # enable GC to save memory
-)
-
 import paddle.fluid as fluid
 
 from ppdet.utils.eval_utils import parse_fetches, eval_run, eval_results, json_eval_results

tools/infer.py

@@ -16,30 +16,16 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-import os
-import glob
-import sys
+import os, sys
 # add python path of PadleDetection to sys.path
 parent_path = os.path.abspath(os.path.join(__file__, *(['..'] * 2)))
 if parent_path not in sys.path:
     sys.path.append(parent_path)
 
+import glob
 import numpy as np
 from PIL import Image
 
-
-def set_paddle_flags(**kwargs):
-    for key, value in kwargs.items():
-        if os.environ.get(key, None) is None:
-            os.environ[key] = str(value)
-
-
-# NOTE(paddle-dev): All of these flags should be set before
-# `import paddle`. Otherwise, it would not take any effect.
-set_paddle_flags(
-    FLAGS_eager_delete_tensor_gb=0,  # enable GC to save memory
-)
-
 from paddle import fluid
 
 from ppdet.core.workspace import load_config, merge_config, create

tools/train.py

@@ -16,31 +16,18 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-import os
+import os, sys
+# add python path of PadleDetection to sys.path
+parent_path = os.path.abspath(os.path.join(__file__, *(['..'] * 2)))
+if parent_path not in sys.path:
+    sys.path.append(parent_path)
+
 import time
 import numpy as np
 import random
 import datetime
 from collections import deque
 from paddle.fluid import profiler
-import sys
-# add python path of PadleDetection to sys.path
-parent_path = os.path.abspath(os.path.join(__file__, *(['..'] * 2)))
-if parent_path not in sys.path:
-    sys.path.append(parent_path)
-
-
-def set_paddle_flags(**kwargs):
-    for key, value in kwargs.items():
-        if os.environ.get(key, None) is None:
-            os.environ[key] = str(value)
-
-
-# NOTE(paddle-dev): All of these flags should be set before
-# `import paddle`. Otherwise, it would not take any effect.
-set_paddle_flags(
-    FLAGS_eager_delete_tensor_gb=0,  # enable GC to save memory
-)
 
 from paddle import fluid
 from paddle.fluid.layers.learning_rate_scheduler import _decay_step_counter