diff --git a/README.md b/README.md index 6755b510b421ddf81eb15a4d538fe4759c569f42..322909ea6f0b36a17a3c5d65ab2d42e30cd13782 100644 --- a/README.md +++ b/README.md @@ -121,8 +121,8 @@ - [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) +- [Objects365 2019 Challenge夺冠模型](docs/featured_model/champion_model/CACascadeRCNN.md) +- [Open Images 2019-Object Detction比赛最佳单模型](docs/featured_model/champion_model/OIDV5_BASELINE_MODEL.md) - [服务器端实用目标检测模型](configs/rcnn_enhance/README.md): V100上速度20FPS时，COCO mAP高达47.8%。 @@ -130,7 +130,7 @@ 本项目的发布受[Apache 2.0 license](LICENSE)许可认证。 ## 版本更新 -v0.2.0版本已经在`02/2020`发布，增加多个模型，升级数据处理模块，拆分YOLOv3的loss，修复已知诸多bug等， +v0.3.0版本已经在`05/2020`发布，增加anchor-free、EfficientDet和YOLOv4等多个模型，推出移动端、服务器端实用高效多个模型，重构预测部署功能，提升易用性，修复已知诸多bug等， 详细内容请参考[版本更新文档](docs/CHANGELOG.md)。 ## 如何贡献代码 diff --git a/README_en.md b/README_en.md index 597d83b08ced91be03be9523180adf9d083469be..dfc351deb89b26b571280fb440f3f038c2597f99 100644 --- a/README_en.md +++ b/README_en.md @@ -134,8 +134,8 @@ The following is the relationship between COCO mAP and FPS on Tesla V100 of repr - [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) +- [Objects365 2019 Challenge champion model](docs/featured_model/champion_model/CACascadeRCNN.md) +- [Best single model of Open Images 2019-Object Detction](docs/featured_model/champion_model/OIDV5_BASELINE_MODEL.md) - [Practical Server-side detection method](configs/rcnn_enhance/README_en.md): Inference speed on single V100 GPU can reach 20FPS when COCO mAP is 47.8%. @@ -143,7 +143,7 @@ The following is the relationship between COCO mAP and FPS on Tesla V100 of repr PaddleDetection is released under the [Apache 2.0 license](LICENSE). ## Updates -v0.2.0 was released at `02/2020`, add some models，Upgrade data processing module, Split YOLOv3's loss, fix many known bugs, etc. +v0.3.0 was released at `05/2020`, add anchor-free, EfficientDet, YOLOv4, etc. Launched mobile and server-side practical and efficient multiple models, refactored predictive deployment functions, and improved ease of use, fix many known bugs, etc. Please refer to [版本更新文档](docs/CHANGELOG.md) for details. ## Contributing diff --git a/configs/mobile/README.md b/configs/mobile/README.md index 6cd8b8b65cd6572831f6e563088acdf206ecd862..d498b50f17cc4b0abcda5ddac41188eebb204964 100755 --- a/configs/mobile/README.md +++ b/configs/mobile/README.md @@ -35,26 +35,28 @@ PaddleDetection目前提供一系列针对移动应用进行优化的模型， - Qualcomm Snapdragon 855 - HiSilicon Kirin 970 - HiSilicon Kirin 980 + - 单CPU线程 (单位： ms) - | | SD625 | SD835 | SD845 | SD855 | Kirin 970 | Kirin 980 | - |------------------|---------|---------|---------|---------|-----------|-----------| - | SSDLite Large | 289.071 | 134.408 | 91.933 | 48.2206 | 144.914 | 55.1186 | - | SSDLite Small | 122.932 | 57.1914 | 41.003 | 22.0694 | 61.5468 | 25.2106 | - | YOLOv3 baseline | 1082.5 | 435.77 | 317.189 | 155.948 | 536.987 | 178.999 | - | YOLOv3 prune | 253.98 | 131.279 | 89.4124 | 48.2856 | 122.732 | 55.8626 | - | Cascade RCNN 320 | 286.526 | 125.635 | 87.404 | 46.184 | 149.179 | 52.9994 | - | Cascade RCNN 640 | 1115.66 | 495.926 | 351.361 | 189.722 | 573.558 | 207.917 | +| | SD625 | SD835 | SD845 | SD855 | Kirin 970 | Kirin 980 | +|------------------|---------|---------|---------|---------|-----------|-----------| +| SSDLite Large | 289.071 | 134.408 | 91.933 | 48.2206 | 144.914 | 55.1186 | +| SSDLite Small | 122.932 | 57.1914 | 41.003 | 22.0694 | 61.5468 | 25.2106 | +| YOLOv3 baseline | 1082.5 | 435.77 | 317.189 | 155.948 | 536.987 | 178.999 | +| YOLOv3 prune | 253.98 | 131.279 | 89.4124 | 48.2856 | 122.732 | 55.8626 | +| Cascade RCNN 320 | 286.526 | 125.635 | 87.404 | 46.184 | 149.179 | 52.9994 | +| Cascade RCNN 640 | 1115.66 | 495.926 | 351.361 | 189.722 | 573.558 | 207.917 | + - 4 CPU线程 (单位： ms) - | | SD625 | SD835 | SD845 | SD855 | Kirin 970 | Kirin 980 | - |------------------|---------|---------|---------|---------|-----------|-----------| - | SSDLite Large | 107.535 | 51.1382 | 34.6392 | 20.4978 | 50.5598 | 24.5318 | - | SSDLite Small | 51.5704 | 24.5156 | 18.5486 | 11.4218 | 24.9946 | 16.7158 | - | YOLOv3 baseline | 413.486 | 184.248 | 133.624 | 75.7354 | 202.263 | 126.435 | - | YOLOv3 prune | 98.5472 | 53.6228 | 34.4306 | 21.3112 | 44.0722 | 31.201 | - | Cascade RCNN 320 | 131.515 | 59.6026 | 39.4338 | 23.5802 | 58.5046 | 36.9486 | - | Cascade RCNN 640 | 473.083 | 224.543 | 156.205 | 100.686 | 231.108 | 138.391 | +| | SD625 | SD835 | SD845 | SD855 | Kirin 970 | Kirin 980 | +|------------------|---------|---------|---------|---------|-----------|-----------| +| SSDLite Large | 107.535 | 51.1382 | 34.6392 | 20.4978 | 50.5598 | 24.5318 | +| SSDLite Small | 51.5704 | 24.5156 | 18.5486 | 11.4218 | 24.9946 | 16.7158 | +| YOLOv3 baseline | 413.486 | 184.248 | 133.624 | 75.7354 | 202.263 | 126.435 | +| YOLOv3 prune | 98.5472 | 53.6228 | 34.4306 | 21.3112 | 44.0722 | 31.201 | +| Cascade RCNN 320 | 131.515 | 59.6026 | 39.4338 | 23.5802 | 58.5046 | 36.9486 | +| Cascade RCNN 640 | 473.083 | 224.543 | 156.205 | 100.686 | 231.108 | 138.391 | ## YOLOv3剪裁说明 diff --git a/configs/mobile/README_en.md b/configs/mobile/README_en.md index 0108e8c67c1a8f08958b8d52ffb4b145c1001798..b2a6f9dd5a1c110967750f80a5c9f48b5c0e023e 100755 --- a/configs/mobile/README_en.md +++ b/configs/mobile/README_en.md @@ -35,6 +35,7 @@ This directory contains models optimized for mobile applications, at present the - Qualcomm Snapdragon 855 - HiSilicon Kirin 970 - HiSilicon Kirin 980 + - With 1 CPU thread (latency numbers are in ms) | | SD625 | SD835 | SD845 | SD855 | Kirin 970 | Kirin 980 | @@ -45,6 +46,7 @@ This directory contains models optimized for mobile applications, at present the | YOLOv3 prune | 253.98 | 131.279 | 89.4124 | 48.2856 | 122.732 | 55.8626 | | Cascade RCNN 320 | 286.526 | 125.635 | 87.404 | 46.184 | 149.179 | 52.9994 | | Cascade RCNN 640 | 1115.66 | 495.926 | 351.361 | 189.722 | 573.558 | 207.917 | + - With 4 CPU threads (latency numbers are in ms) | | SD625 | SD835 | SD845 | SD855 | Kirin 970 | Kirin 980 | diff --git a/configs/rcnn_enhance/README.md b/configs/rcnn_enhance/README.md index 44f9ba043419a7d1cd00241ba035f5027bba49c5..08428fbcef4ea14eabb701ff888e1bff2c9bf2b7 100644 --- a/configs/rcnn_enhance/README.md +++ b/configs/rcnn_enhance/README.md @@ -1,4 +1,4 @@ -## 服务器端实用目标检测方案(Practical Server-side detection, PSS-DET) +## 服务器端实用目标检测方案 ### 简介 @@ -28,7 +28,7 @@ > 这里为了更方便地对比，统一将V100的预测耗时乘以1.2倍，近似转化为Titan V的预测耗时。 -## 模型库 +### 模型库 | 骨架网络 | 网络类型 | 每张GPU图片个数 | 学习率策略 |推理时间(fps) | Box AP | Mask AP | 下载 | 配置文件 | | :---------------------- | :-------------: | :-------: | :-----: | :------------: | :----: | :-----: | :-------------: | :-----: | diff --git a/configs/yolo/README.md b/configs/yolov4/README.md similarity index 93% rename from configs/yolo/README.md rename to configs/yolov4/README.md index acf5cfbc74a8100b76d4aae92e31c66bf604675f..98a53258036519737068144a6fffa75ee63adcff 100644 --- a/configs/yolo/README.md +++ b/configs/yolov4/README.md @@ -1,4 +1,4 @@ -# YOLO v4 +# YOLO v4 模型 ## 内容 - [简介](#简介) @@ -25,8 +25,8 @@ | | GPU个数 | 测试集 | 骨干网络 | 精度 | 模型下载 | 配置文件 | |:------------------------:|:-------:|:------:|:--------------------------:|:------------------------:| :---------:| :-----: | -| YOLO v4 | - |test-dev2019 | CSPDarkNet53 | 43.5 |[下载链接](https://paddlemodels.bj.bcebos.com/object_detection/yolov4_cspdarknet.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/yolo/yolov4_cspdarknet.yml) | -| YOLO v4 VOC | 2 | VOC2007 | CSPDarkNet53 | 85.5 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/yolov4_cspdarknet_voc.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/yolo/yolov4_cspdarknet_voc.yml) | +| YOLO v4 | - |test-dev2019 | CSPDarkNet53 | 43.5 |[下载链接](https://paddlemodels.bj.bcebos.com/object_detection/yolov4_cspdarknet.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/yolov4/yolov4_cspdarknet.yml) | +| YOLO v4 VOC | 2 | VOC2007 | CSPDarkNet53 | 85.5 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/yolov4_cspdarknet_voc.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/yolov4/yolov4_cspdarknet_voc.yml) | **注意:** diff --git a/configs/yolo/yolov4_cspdarknet.yml b/configs/yolov4/yolov4_cspdarknet.yml similarity index 100% rename from configs/yolo/yolov4_cspdarknet.yml rename to configs/yolov4/yolov4_cspdarknet.yml diff --git a/configs/yolo/yolov4_cspdarknet_voc.yml b/configs/yolov4/yolov4_cspdarknet_voc.yml similarity index 100% rename from configs/yolo/yolov4_cspdarknet_voc.yml rename to configs/yolov4/yolov4_cspdarknet_voc.yml diff --git a/deploy/cpp/README.md b/deploy/cpp/README.md index 2bc464daa7a68aeaf25a8b03c9e18a1d0b5f20c4..d7e55d1a7bf061f00a85255bde8de84b8e5dc66b 100644 --- a/deploy/cpp/README.md +++ b/deploy/cpp/README.md @@ -1,6 +1,6 @@ -# PaddleDetection C++预测部署方案 +# C++端预测部署 -## 本文档结构 +## 本教程结构 [1.说明](#1说明) diff --git a/deploy/python/README.md b/deploy/python/README.md index e8252796c3bebb33a9d0a5fffd32d920a0e7d00a..5961ad4975abb7530775dec1328b24d6d9a72392 100644 --- a/deploy/python/README.md +++ b/deploy/python/README.md @@ -1,4 +1,4 @@ -## PaddleDetection Python 预测部署方案 +# Python端预测部署 本篇教程使用AnalysisPredictor对[导出模型](../../docs/advanced_tutorials/deploy/EXPORT_MODEL.md)进行高性能预测。 在PaddlePaddle中预测引擎和训练引擎底层有着不同的优化方法, 下面列出了两种不同的预测方式。Executor同时支持训练和预测，AnalysisPredictor则专门针对推理进行了优化，是基于[C++预测库](https://www.paddlepaddle.org.cn/documentation/docs/zh/advanced_guide/inference_deployment/inference/native_infer.html)的Python接口，该引擎可以对模型进行多项图优化，减少不必要的内存拷贝。如果用户在部署已训练模型的过程中对性能有较高的要求，我们提供了独立于PaddleDetection的预测脚本，方便用户直接集成部署。 @@ -19,6 +19,7 @@ PaddleDetection在训练过程包括网络的前向和优化器相关参数， 导出后目录下，包括`__model__`，`__params__`和`infer_cfg.yml`三个文件。 ## 2. 基于python的预测 + ### 2.1 安装依赖 - `PaddlePaddle`的安装: 请点击[官方安装文档](https://paddlepaddle.org.cn/install/quick) 选择适合的方式，版本为1.7以上即可 diff --git a/docs/CHANGELOG.md b/docs/CHANGELOG.md index 9f4eb687afa62607cf065eac9fbd7c720e6cfeba..f8f2c2455276f9212249014775a0a935fc821717 100644 --- a/docs/CHANGELOG.md +++ b/docs/CHANGELOG.md @@ -1,5 +1,34 @@ # 版本更新信息 +## 最新版本信息 + +### v0.3.0(05/2020) + - 模型丰富度提升： + - 添加Efficientdet-D0模型，速度与精度优于竞品。 + - 添加anchor-free模型FCOS，精度优于竞品。 + - 新增yolov4预测模型，精度对齐竞品；新增yolov4在pascal voc数据集上finetune模型，精度达到85.5%。 + - YOLOv3新增MobileNetV3骨干网络，COCO数据集精度达到31.6%。 + - 添加anchor-free模型CornernetSqueeze，精度优于竞品， 优化模型的COCO数据集精度38.2%, +3.7%，速度较yolo_v3 darknet快5%。 + - 添加服务器端实用目标检测模型cascade_rcnn_resnet50_vd_fpn_dcn，速度与精度优于竞品EfficientDet。 + + - 移动端推出3种模型： + - SSDLite系列模型：ssdlite-mobilenet_v3 large模型与ssdlite-mobilenet_v3 small模型，精度优于竞品。ssdlite-mobilenet_v1模型，精度优于竞品。 + - yolo v3：yolov3_mobilenet_v3裁剪模型，速度和精度均领先于竞品的SSDLite模型。 + - Faster RCNN：cascade_rcnn_mobilenet_v3 large_fpn推出输入图像分别为320x320和640x640的模型，速度与精度具有较高性价比。 + + - 预测部署重构： + - 新增Python预测部署流程，支持RCNN，YOLO，SSD，RetinaNet，人脸系列模型，支持视频预测。 + - 重构C++预测部署，提高易用性。 + + - 易用性提升及功能组件： + - 增加AutoAugment数据增强。 + - 升级检测库文档结构。 + - 支持迁移学习自动进行shape匹配。 + - 优化mask分支评估阶段内存占用。 + + +## 历史版本信息 + ### v0.2.0(02/2020) - 新增模型: - 新增基于CBResNet模型。 diff --git a/docs/tutorials/FAQ.md b/docs/FAQ.md similarity index 99% rename from docs/tutorials/FAQ.md rename to docs/FAQ.md index c7d7124e004c47f14c703fbb32954b40322bde74..63811d846d8adf761258705f9de5c3e9a82b7e78 100644 --- a/docs/tutorials/FAQ.md +++ b/docs/FAQ.md @@ -1,4 +1,4 @@ -## FAQ +## FAQ(常见问题) **Q:** 为什么我使用单GPU训练loss会出`NaN`?
**A:** 默认学习率是适配多GPU训练(8x GPU)，若使用单GPU训练，须对应调整学习率（例如，除以8）。 diff --git a/docs/advanced_tutorials/deploy/index.rst b/docs/advanced_tutorials/deploy/index.rst index a927e8b9cfd34b9443983d3af459fd03a465cfaf..296cf744911d00628fb3dced2db2e6667c05c0ad 100644 --- a/docs/advanced_tutorials/deploy/index.rst +++ b/docs/advanced_tutorials/deploy/index.rst @@ -1,4 +1,4 @@ -推理部署 +推理部署教程 =========================================== .. toctree:: @@ -8,3 +8,5 @@ DEPLOY_PY.md DEPLOY_CPP.md BENCHMARK_INFER_cn.md + +.. note:: 文中超链接以GitHub中展示为准，如出现超链接无法访问，请点击网页右上角「Edit on github」查看源文件进行索引，有任何问题欢迎在 `GitHub `_ 上提issue。 diff --git a/docs/advanced_tutorials/index.rst b/docs/advanced_tutorials/index.rst index 5f1f14905a9daaf0518b667abf447386741a8e04..95c215f5cfc40b3271b67735c5313c5df4703f49 100644 --- a/docs/advanced_tutorials/index.rst +++ b/docs/advanced_tutorials/index.rst @@ -1,4 +1,4 @@ -高级使用教程 +进阶使用教程 =========================================== .. toctree:: @@ -10,3 +10,5 @@ CONFIG_cn.md TRANSFER_LEARNING_cn.md slim/index + +.. note:: 文中超链接以GitHub中展示为准，如出现超链接无法访问，请点击网页右上角「Edit on github」查看源文件进行索引，有任何问题欢迎在 `GitHub `_ 上提issue。 diff --git a/docs/advanced_tutorials/slim/index.rst b/docs/advanced_tutorials/slim/index.rst index a8af21d6f12cb4e5d33dc095dab06dd4113023cc..614b168b17f433aa8c93b113cccb40983c2c16e4 100644 --- a/docs/advanced_tutorials/slim/index.rst +++ b/docs/advanced_tutorials/slim/index.rst @@ -9,3 +9,5 @@ quantization/index nas/index prune/index + +.. note:: 文中超链接以GitHub中展示为准，如出现超链接无法访问，请点击网页右上角「Edit on github」查看源文件进行索引，有任何问题欢迎在 `GitHub `_ 上提issue。 diff --git a/docs/featured_model/ANCHOR_FREE_DETECTION.md b/docs/featured_model/ANCHOR_FREE_DETECTION.md deleted file mode 100644 index 2563896f2ced185838873b2e9ae520efd8f990f5..0000000000000000000000000000000000000000 --- a/docs/featured_model/ANCHOR_FREE_DETECTION.md +++ /dev/null @@ -1 +0,0 @@ -**文档教程请参考：** [ACHOR\_FREE\_DETECTION.md](../../configs/anchor_free/README.md)
diff --git a/docs/featured_model/ANCHOR_FREE_DETECTION.md b/docs/featured_model/ANCHOR_FREE_DETECTION.md new file mode 120000 index 0000000000000000000000000000000000000000..5c1a8fb105254339c49319a5c38cfac924891eb4 --- /dev/null +++ b/docs/featured_model/ANCHOR_FREE_DETECTION.md @@ -0,0 +1 @@ +../../configs/anchor_free/README.md \ No newline at end of file diff --git a/docs/featured_model/FACE_DETECTION.md b/docs/featured_model/FACE_DETECTION.md index eed2c4439d9d2e89168868f7905793c96a089135..efa4041ac6f2de4786277393cc3760f240eab21e 100644 --- a/docs/featured_model/FACE_DETECTION.md +++ b/docs/featured_model/FACE_DETECTION.md @@ -1,5 +1,5 @@ [English](FACE_DETECTION_en.md) | 简体中文 -# FaceDetection +# 人脸检测模型 ## 内容 - [简介](#简介) diff --git a/docs/featured_model/MOBILE_SIDE.md b/docs/featured_model/MOBILE_SIDE.md new file mode 120000 index 0000000000000000000000000000000000000000..9d294f2c0228cdefd41c5d7936c81566d0b2f95e --- /dev/null +++ b/docs/featured_model/MOBILE_SIDE.md @@ -0,0 +1 @@ +../../configs/mobile/README.md \ No newline at end of file diff --git a/docs/featured_model/SERVER_SIDE.md b/docs/featured_model/SERVER_SIDE.md new file mode 120000 index 0000000000000000000000000000000000000000..ca9cb6863f3829ae0a0f69aaeccc3cdd19308530 --- /dev/null +++ b/docs/featured_model/SERVER_SIDE.md @@ -0,0 +1 @@ +../../configs/rcnn_enhance/README.md \ No newline at end of file diff --git a/docs/featured_model/YOLO_V4.md b/docs/featured_model/YOLO_V4.md new file mode 120000 index 0000000000000000000000000000000000000000..192f0ebb5fe81b2236dcec4c4991af211a0030c1 --- /dev/null +++ b/docs/featured_model/YOLO_V4.md @@ -0,0 +1 @@ +../../configs/yolov4/README.md \ No newline at end of file diff --git a/docs/featured_model/CACascadeRCNN.md b/docs/featured_model/champion_model/CACascadeRCNN.md similarity index 96% rename from docs/featured_model/CACascadeRCNN.md rename to docs/featured_model/champion_model/CACascadeRCNN.md index ec438e8cd3ab3f2d41b6784976bf3007128604cd..f1f19dc8b2a95a928380cec0001d98c1c724e58c 100644 --- a/docs/featured_model/CACascadeRCNN.md +++ b/docs/featured_model/champion_model/CACascadeRCNN.md @@ -2,13 +2,13 @@ ## 简介 CACascade RCNN是百度视觉技术部在Objects365 2019 Challenge上夺冠的最佳单模型之一，Objects365是在通用物体检测领域的一个全新的数据集，旨在促进对自然场景不同对象的检测研究。Objects365在63万张图像上标注了365个对象类，训练集中共有超过1000万个边界框。这里放出的是Full Track任务中最好的单模型之一。 -![](../images/obj365_gt.png) +![](../../images/obj365_gt.png) ## 方法描述 针对大规模物体检测算法的特点，我们提出了一种基于图片包含物体类别的数量的采样方式（Class Aware Sampling）。基于这种方式进行训练模型可以在更短的时间使模型收敛到更好的效果。 -![](../images/cas.png) +![](../../images/cas.png) 本次公布的最好单模型是一个基于Cascade RCNN的两阶段检测模型，在此基础上将Backbone替换为更加强大的SENet154模型，Deformable Conv模块以及更复杂二阶段网络结构，针对BatchSize比较小的情况增加了Group Normalization操作并同时使用了多尺度训练，最终达到了非常理想的效果。预训练模型先后分别在ImageNet和COCO数据集上进行了训练，其中在COCO数据集上训练时增加了Mask分支，其余结构与CACascade RCNN相同， 会在启动训练时自动下载。 @@ -42,4 +42,4 @@ python tools/train.py -c configs/obj365/cascade_rcnn_dcnv2_se154_vd_fpn_gn_cas.y ## 模型效果 -![](../images/obj365_pred.png) +![](../../images/obj365_pred.png) diff --git a/docs/featured_model/OIDV5_BASELINE_MODEL.md b/docs/featured_model/champion_model/OIDV5_BASELINE_MODEL.md similarity index 97% rename from docs/featured_model/OIDV5_BASELINE_MODEL.md rename to docs/featured_model/champion_model/OIDV5_BASELINE_MODEL.md index 1bfcabdba1f0503ce2131f5efd021242b5a2e19e..30a374710a16a11e1a832168475329e8dc95bd13 100644 --- a/docs/featured_model/OIDV5_BASELINE_MODEL.md +++ b/docs/featured_model/champion_model/OIDV5_BASELINE_MODEL.md @@ -2,12 +2,12 @@ ## 简介 CascadeCA RCNN是百度视觉技术部在Google AI Open Images 2019-Object Detction比赛中的最佳单模型，该单模型助力团队在500多参数队伍中取得第二名。Open Images Dataset V5(OIDV5)包含500个类别、173W训练图像和超过1400W个标注边框，是目前已知规模最大的目标检测公开数据集，数据集地址：[https://storage.googleapis.com/openimages/web/index.html](https://storage.googleapis.com/openimages/web/index.html)。团队在比赛中的技术方案报告地址：[https://arxiv.org/pdf/1911.07171.pdf](https://arxiv.org/pdf/1911.07171.pdf) -![](../images/oidv5_gt.png) +![](../../images/oidv5_gt.png) ## 方法描述 该模型结合了当前较优的检测方法。具体地，它将ResNet200-vd作为检测模型的骨干网络，其imagenet分类预训练模型可以在[这里](https://github.com/PaddlePaddle/models/blob/develop/PaddleCV/image_classification/README_en.md)下载；结合了CascadeCA RCNN、Feature Pyramid Networks、Non-local、Deformable V2等方法。在这里需要注意的是，标准的CascadeRCNN是只预测2个框（前景和背景，使用得分信息去判断最终前景所属的类别），而该模型对每个类别都单独预测了一个框（Cascade Class Aware）。最终模型框图如下图所示。 -![](../images/oidv5_model_framework.png) +![](../../images/oidv5_model_framework.png) 由于OIDV5的类别不均衡现象比较严重，在训练时采用了动态采样的策略去选择样本并进行训练；多尺度训练被用于解决边框面积范围太大的情况；此外，团队使用Libra loss替代Smooth L1 loss，来计算预测框的loss；在预测时，使用SoftNMS方法进行后处理，保证更多的框可以被召回。 @@ -49,4 +49,4 @@ python -u tools/infer.py -c configs/oidv5/cascade_rcnn_cls_aware_r200_vd_fpn_dcn ## 模型检测效果 -![](../images/oidv5_pred.jpg) +![](../../images/oidv5_pred.jpg) diff --git a/docs/featured_model/champion_model/index.rst b/docs/featured_model/champion_model/index.rst new file mode 100644 index 0000000000000000000000000000000000000000..d6d9c7374c33839e01e741580938c5d783b7a226 --- /dev/null +++ b/docs/featured_model/champion_model/index.rst @@ -0,0 +1,10 @@ +竞赛冠军模型 +=========================================== + +.. toctree:: + :maxdepth: 2 + + CACascadeRCNN.md + OIDV5_BASELINE_MODEL.md + +.. note:: 文中超链接以GitHub中展示为准，如出现超链接无法访问，请点击网页右上角「Edit on github」查看源文件进行索引，有任何问题欢迎在 `GitHub `_ 上提issue。 diff --git a/docs/featured_model/index.rst b/docs/featured_model/index.rst index 299d62af36aba163ce6627358ac6095a451272a1..44f55853da9cb18d51c2609b33eeb9329bbbb722 100644 --- a/docs/featured_model/index.rst +++ b/docs/featured_model/index.rst @@ -4,8 +4,13 @@ .. toctree:: :maxdepth: 2 - FACE_DETECTION.md YOLOv3_ENHANCEMENT.md - CACascadeRCNN.md - OIDV5_BASELINE_MODEL.md + MOBILE_SIDE.md + SERVER_SIDE.md + ANCHOR_FREE_DETECTION.md + YOLO_V4.md + champion_model/index.rst + FACE_DETECTION.md CONTRIB_cn.md + +.. note:: 文中超链接以GitHub中展示为准，如出现超链接无法访问，请点击网页右上角「Edit on github」查看源文件进行索引，有任何问题欢迎在 `GitHub `_ 上提issue。 diff --git a/docs/index.rst b/docs/index.rst index e2ac7dfdb1841f19254f4cfa4baebed0dce48371..89cae2aec62033874fbd0975e6b45e166cc624e9 100644 --- a/docs/index.rst +++ b/docs/index.rst @@ -9,3 +9,6 @@ featured_model/index MODEL_ZOO_cn.md CHANGELOG.md + FAQ.md + +.. note:: 文中超链接以GitHub中展示为准，如出现超链接无法访问，请点击网页右上角「Edit on github」查看源文件进行索引，有任何问题欢迎在 `GitHub `_ 上提issue。 diff --git a/docs/tutorials/index.rst b/docs/tutorials/index.rst index 0385863a84faea47bd5ae64709e0ebb700374259..2c68837aa7a9ba0d976b937d3c21843c6d9c6d58 100644 --- a/docs/tutorials/index.rst +++ b/docs/tutorials/index.rst @@ -1,4 +1,4 @@ -初级使用教程 +入门使用教程 =========================================== .. toctree:: @@ -7,4 +7,5 @@ INSTALL_cn.md QUICK_STARTED_cn.md GETTING_STARTED_cn.md - FAQ.md + +.. note:: 文中超链接以GitHub中展示为准，如出现超链接无法访问，请点击网页右上角「Edit on github」查看源文件进行索引，有任何问题欢迎在 `GitHub `_ 上提issue。