diff --git a/README_cn.md b/README_cn.md index 5a902ee71f5cb34dc482ad1bc34468f3f0e0bd8b..7386cbcaa3719271ef96e0d86d996f5bd6869812 100644 --- a/README_cn.md +++ b/README_cn.md @@ -4,7 +4,7 @@ # 产品动态 - 2021.11.03: 发布[release/2.3版本](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.3)，发布**轻量级检测特色模型**⚡[PP-PicoDet](configs/picodet)，发布**轻量级关键点特色模型**⚡[PP-TinyPose](configs/keypoint/tiny_pose)。新增[Swin Transformer](configs/faster_rcnn)，[TOOD](configs/tood)，[GFL](configs/gfl)目标检测模型。发布[Sniper](configs/sniper)小目标检测优化模型，发布针对EdgeBoard优化[PP-YOLO-EB](configs/ppyolo)模型。新增轻量化关键点模型[Lite HRNet](configs/keypoint)关键点模型并支持Paddle Lite部署。 -- 2021.08.10: 发布[release/2.2版本]((https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.2))，发布Transformer检测系列模型，包括[DETR](configs/detr), [Deformable DETR](configs/deformable_detr), [Sparse RCNN](configs/sparse_rcnn)。新增Dark HRNet关键点模型和MPII数据集[关键点模型](configs/keypoint)，新增[人头](configs/mot/headtracking21)、[车辆](configs/mot/vehicle)跟踪垂类模型。 +- 2021.08.10: 发布[release/2.2版本](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.2)，发布Transformer检测系列模型，包括[DETR](configs/detr), [Deformable DETR](configs/deformable_detr), [Sparse RCNN](configs/sparse_rcnn)。新增Dark HRNet关键点模型和MPII数据集[关键点模型](configs/keypoint)，新增[人头](configs/mot/headtracking21)、[车辆](configs/mot/vehicle)跟踪垂类模型。 - 2021.05.20: 发布[release/2.1版本](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.1)，新增[关键点检测](configs/keypoint)，模型包括HigherHRNet，HRNet。新增[多目标跟踪](configs/mot)能力，模型包括DeepSORT，JDE，FairMOT。发布PPYOLO系列模型压缩模型，新增[ONNX模型导出教程](deploy/EXPORT_ONNX_MODEL.md)。 # 简介 @@ -13,13 +13,8 @@ PaddleDetection为基于飞桨PaddlePaddle的端到端目标检测套件，提 ### PaddleDetection提供了目标检测、实例分割、多目标跟踪、关键点检测等多种能力 -
- - -
-
- - +
+
@@ -33,7 +28,7 @@ PaddleDetection为基于飞桨PaddlePaddle的端到端目标检测套件，提 ### 套件结构概览 - +
@@ -186,6 +214,16 @@ PaddleDetection为基于飞桨PaddlePaddle的端到端目标检测套件，提 - `PP-YOLO v2`是对`PP-YOLO`模型的进一步优化，在COCO数据集精度49.5%，Tesla V100预测速度68.9FPS - 图中模型均可在[模型库](#模型库)中获取 +各移动端模型在COCO数据集上精度mAP和高通骁龙865处理器上预测速度(FPS)对比图。 + +
+ +
+ +**说明：** +- 测试数据均使用高通骁龙865(4\*A77 + 4\*A55)处理器batch size为1, 开启4线程测试，测试使用NCNN预测库，测试脚本见[MobileDetBenchmark](https://github.com/JiweiMaster/MobileDetBenchmark) +- [PP-PicoDet](configs/picodet)及[PP-YOLO-Tiny](configs/ppyolo)为PaddleDetection自研模型，其余模型PaddleDetection暂未提供 + ## 文档教程 ### 入门教程 diff --git a/README_en.md b/README_en.md index 8bd0c96c44cda90401728ff12a31dbf6afd203e1..19e2233c611e0225335925b9f0b0f1c15771bb8c 100644 --- a/README_en.md +++ b/README_en.md @@ -3,8 +3,8 @@ English | [简体中文](README_cn.md) # Product news -- 2021.11.03: Release [release/2.3]((https://github.com/PaddlePaddle/Paddleetection/tree/release/2.3) version. Release mobile object detection model ⚡[PP-PicoDet](configs/picodet), mobile keypoint detection model ⚡[PP-TinyPose](configs/keypoint/tiny_pose). Release object detection models, including [Swin-Transformer](configs/faster_rcnn), [TOOD](configs/tood), [GFL](configs/gfl), release [Sniper](configs/sniper) tiny object detection models and optimized [PP-YOLO-EB](configs/ppyolo) model for EdgeBoard. Release mobile keypoint detection model [Lite HRNet](configs/keypoint). -- 2021.08.10: Release [release/2.2]((https://github.com/PaddlePaddle/Paddleetection/tree/release/2.2) version. Release Transformer object detection models, including [DETR](configs/detr), [Deformable DETR](configs/deformable_detr), [Sparse RCNN](configs/sparse_rcnn). Release [keypoint detection](configs/keypoint) models, including DarkHRNet and model trained on MPII dataset. Release [head-tracking](configs/mot/headtracking21) and [vehicle-tracking](configs/mot/vehicle) multi-object tracking models. +- 2021.11.03: Release [release/2.3](https://github.com/PaddlePaddle/Paddleetection/tree/release/2.3) version. Release mobile object detection model ⚡[PP-PicoDet](configs/picodet), mobile keypoint detection model ⚡[PP-TinyPose](configs/keypoint/tiny_pose). Release object detection models, including [Swin-Transformer](configs/faster_rcnn), [TOOD](configs/tood), [GFL](configs/gfl), release [Sniper](configs/sniper) tiny object detection models and optimized [PP-YOLO-EB](configs/ppyolo) model for EdgeBoard. Release mobile keypoint detection model [Lite HRNet](configs/keypoint). +- 2021.08.10: Release [release/2.2](https://github.com/PaddlePaddle/Paddleetection/tree/release/2.2) version. Release Transformer object detection models, including [DETR](configs/detr), [Deformable DETR](configs/deformable_detr), [Sparse RCNN](configs/sparse_rcnn). Release [keypoint detection](configs/keypoint) models, including DarkHRNet and model trained on MPII dataset. Release [head-tracking](configs/mot/headtracking21) and [vehicle-tracking](configs/mot/vehicle) multi-object tracking models. - 2021.05.20: Release [release/2.1]((https://github.com/PaddlePaddle/Paddleetection/tree/release/2.1) version. Release [Keypoint Detection](configs/keypoint), including HigherHRNet and HRNet, [Multi-Object Tracking](configs/mot), including DeepSORT，JDE and FairMOT. Release model compression for PPYOLO series models.Update documents such as [EXPORT ONNX MODEL](deploy/EXPORT_ONNX_MODEL.md). @@ -14,13 +14,8 @@ PaddleDetection is an end-to-end object detection development kit based on Paddl ### PaddleDetection provides image processing capabilities such as object detection, instance segmentation, multi-object tracking, keypoint detection and etc. -
- - -
-
- - +
+
@@ -40,7 +35,7 @@ Based on the high performance core of PaddlePaddle, advantages of training speed #### Overview of Kit Structures -
@@ -66,7 +61,7 @@ PaddleDetection为基于飞桨PaddlePaddle的端到端目标检测套件，提
  • RetinaNet
  • YOLOv3
  • YOLOv4
    • -
  • PP-YOLO
    • +
  • PP-YOLOv1/v2/Tiny
  • SSD
    • @@ -75,6 +70,7 @@ PaddleDetection为基于飞桨PaddlePaddle的端到端目标检测套件，提
  • CornerNet-Squeeze
  • FCOS
  • TTFNet
    • +
  • PicoDet
      • @@ -92,6 +88,29 @@ PaddleDetection为基于飞桨PaddlePaddle的端到端目标检测套件，提
    • BlazeFace-NAS
    +
      +
    • Transformer
      • +
        +
      • DETR/Deformable DETR
        • +
      • Sparse RCNN
        • +
      • Swin Transformer
        • +
      +
    +
      +
    • Multi-Object-Tracking
      • +
        +
      • JDE
        • +
      • FairMOT
        • +
      • DeepSort
        • +
      +
    +
      +
    • KeyPoint-Detection
      • +
        +
      • HRNet
        • +
      • HigherHRNet
        • +
      +
      @@ -109,6 +128,7 @@ PaddleDetection为基于飞桨PaddlePaddle的端到端目标检测套件，提
    • MobileNetv1/v3
    • GhostNet
    • Efficientnet
      • +
    • BlazeNet
    		 @@ -120,6 +140,11 @@ PaddleDetection为基于飞桨PaddlePaddle的端到端目标检测套件，提
  • Non-local
    • +
    • KeyPoint
      • +
        +
      • DarkPose
        • +
      +
    • FPN
      • BiFPN
        • @@ -151,15 +176,18 @@ PaddleDetection为基于飞桨PaddlePaddle的端到端目标检测套件，提
    • Resize
      • +
    • Lighting
    • Flipping
    • Expand
    • Crop
    • Color Distort
    • Random Erasing
    • Mixup
      • +
    • Mosaic
    • Cutmix
    • Grid Mask
    • Auto Augment
      • +
    • Random Perspective
    +
    @@ -178,24 +206,35 @@ Based on the high performance core of PaddlePaddle, advantages of training speed
    @@ -65,7 +60,7 @@ Based on the high performance core of PaddlePaddle, advantages of training speed
  • Cascade-RCNN
  • Libra RCNN
  • Hybrid Task RCNN
    • -
  • PSS-Det RCNN
    • +
  • PSS-Det
    • One-Stage Detection
      • @@ -73,7 +68,7 @@ Based on the high performance core of PaddlePaddle, advantages of training speed
    • RetinaNet
    • YOLOv3
    • YOLOv4
      • -
    • PP-YOLO
      • +
    • PP-YOLOv1/v2/Tiny
    • SSD
    @@ -82,6 +77,7 @@ Based on the high performance core of PaddlePaddle, advantages of training speed
  • CornerNet-Squeeze
  • FCOS
  • TTFNet
    • +
  • PicoDet
      • @@ -92,13 +88,36 @@ Based on the high performance core of PaddlePaddle, advantages of training speed
      -
    • Face-Detection
      • +
    • Face-Detction
      • FaceBoxes
      • BlazeFace
      • BlazeFace-NAS
    +
      +
    • Transformer
      • +
        +
      • DETR/Deformable DETR
        • +
      • Sparse RCNN
        • +
      • Swin Transformer
        • +
      +
    +
      +
    • Multi-Object-Tracking
      • +
        +
      • JDE
        • +
      • FairMOT
        • +
      • DeepSort
        • +
      +
    +
      +
    • KeyPoint-Detection
      • +
        +
      • HRNet
        • +
      • HigherHRNet
        • +
      +
      @@ -116,6 +135,7 @@ Based on the high performance core of PaddlePaddle, advantages of training speed
    • MobileNetv1/v3
    • GhostNet
    • Efficientnet
      • +
    • BlazeNet
    		 @@ -127,6 +147,11 @@ Based on the high performance core of PaddlePaddle, advantages of training speed
  • Non-local
    • +
    • KeyPoint
      • +
        +
      • DarkPose
        • +
      +
    • FPN
      • BiFPN
        • @@ -158,15 +183,18 @@ Based on the high performance core of PaddlePaddle, advantages of training speed
    • Resize
      • +
    • Lighting
    • Flipping
    • Expand
    • Crop
    • Color Distort
    • Random Erasing
    • Mixup
      • +
    • Mosaic
    • Cutmix
    • Grid Mask
    • Auto Augment
      • +
    • Random Perspective
    #### Overview of Model Performance -The relationship between COCO mAP and FPS on Tesla V100 of representative models of each architectures and backbones. + +The relationship between COCO mAP and FPS on Tesla V100 of representative models of each server side architectures and backbones.
    -
    +
    -**NOTE:** + **NOTE:** -- `CBResNet stands` for `Cascade-Faster-RCNN-CBResNet200vd-FPN`, which has highest mAP on COCO as 53.3% + - `CBResNet stands` for `Cascade-Faster-RCNN-CBResNet200vd-FPN`, which has highest mAP on COCO as 53.3% -- `Cascade-Faster-RCNN` stands for `Cascade-Faster-RCNN-ResNet50vd-DCN`, which has been optimized to 20 FPS inference speed when COCO mAP as 47.8% in PaddleDetection models + - `Cascade-Faster-RCNN` stands for `Cascade-Faster-RCNN-ResNet50vd-DCN`, which has been optimized to 20 FPS inference speed when COCO mAP as 47.8% in PaddleDetection models -- `PP-YOLO` achieves mAP of 45.9% on COCO and 72.9FPS on Tesla V100. Both precision and speed surpass [YOLOv4](https://arxiv.org/abs/2004.10934) + - `PP-YOLO` achieves mAP of 45.9% on COCO and 72.9FPS on Tesla V100. Both precision and speed surpass [YOLOv4](https://arxiv.org/abs/2004.10934) -- `PP-YOLO v2` is optimized version of `PP-YOLO` which has mAP of 49.5% and 68.9FPS on Tesla V100 + - `PP-YOLO v2` is optimized version of `PP-YOLO` which has mAP of 49.5% and 68.9FPS on Tesla V100 -- All these models can be get in [Model Zoo](#Model-Zoo) + - All these models can be get in [Model Zoo](#ModelZoo) + +The relationship between COCO mAP and FPS on Qualcomm Snapdragon 865 of representative mobile side models. + +
    + +
    + +**NOTE:** +- All data tested on Qualcomm Snapdragon 865(4\*A77 + 4\*A55) processor with batch size of 1 and CPU threads of 4, and use NCNN library in testing, benchmark scripts is publiced at [MobileDetBenchmark](https://github.com/JiweiMaster/MobileDetBenchmark) +- [PP-PicoDet](configs/picodet) and [PP-YOLO-Tiny](configs/ppyolo) are developed and released by PaddleDetection, other models are not provided in PaddleDetection. ## Tutorials diff --git a/docs/images/det.jpg b/docs/images/det.jpg deleted file mode 100644 index b8acf0a7a8002b57a9ef9b713a14a8924ed9160c..0000000000000000000000000000000000000000 Binary files a/docs/images/det.jpg and /dev/null differ diff --git a/docs/images/ins.jpg b/docs/images/ins.jpg deleted file mode 100644 index 96301ad83a3d997dd574d6d52dad328f30663dd4..0000000000000000000000000000000000000000 Binary files a/docs/images/ins.jpg and /dev/null differ diff --git a/docs/images/mobile_fps_map.png b/docs/images/mobile_fps_map.png new file mode 100644 index 0000000000000000000000000000000000000000..2b31508332710042406ab046529148d82a0581e8 Binary files /dev/null and b/docs/images/mobile_fps_map.png differ diff --git a/docs/images/pose.gif b/docs/images/pose.gif deleted file mode 100644 index 605c7506255cc9a18669629a995ddfbe099b3f54..0000000000000000000000000000000000000000 Binary files a/docs/images/pose.gif and /dev/null differ diff --git a/docs/images/mot.gif b/docs/images/ppdet.gif similarity index 57% rename from docs/images/mot.gif rename to docs/images/ppdet.gif index 60c169915c6045627210361d6a4450e413c7bdf1..9539637070da504ead2d5870eb512b438bcd5c62 100644 Binary files a/docs/images/mot.gif and b/docs/images/ppdet.gif differ