From e98e5f36c638a4e6304f8288192c920317528bbf Mon Sep 17 00:00:00 2001 From: Feng Ni Date: Fri, 17 Jun 2022 09:48:56 +0800 Subject: [PATCH] [cherry-pick] update pphuman mot doc and add ppyoloe-s (#6211) * add ppyoloe-s, update mot doc, test=document_fix * fix doc typo, test=document_fix --- deploy/pphuman/README.md | 6 ++++-- deploy/pphuman/README_en.md | 6 ++++-- deploy/pphuman/config/tracker_config.yml | 13 +------------ deploy/pphuman/docs/mot.md | 9 +++++---- deploy/pphuman/docs/mot_en.md | 7 ++++--- 5 files changed, 18 insertions(+), 23 deletions(-) diff --git a/deploy/pphuman/README.md b/deploy/pphuman/README.md index 0496f3f46..864c30afb 100644 --- a/deploy/pphuman/README.md +++ b/deploy/pphuman/README.md @@ -42,8 +42,10 @@ PP-Human提供了目标检测、属性识别、行为识别、ReID预训练模 | 任务 | 适用场景 | 精度 | 预测速度（ms） | 模型权重 | 预测部署模型 | | :---------: |:---------: |:--------------- | :-------: | :------: | :------: | -| 目标检测 | 图片输入 | mAP: 56.3 | 28.0ms |[下载链接](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.pdparams) |[下载链接](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.zip) | -| 目标跟踪 | 视频输入 | MOTA: 72.0 | 33.1ms |[下载链接](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.pdparams) |[下载链接](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.zip) | +| 目标检测(高精度) | 图片输入 | mAP: 56.6 | 28.0ms |[下载链接](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.pdparams) |[下载链接](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.zip) | +| 目标检测(轻量级) | 图片输入 | mAP: 53.2 | 22.1ms |[下载链接](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_s_36e_pipeline.pdparams) |[下载链接](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_s_36e_pipeline.zip) | +| 目标跟踪(高精度) | 视频输入 | MOTA: 79.5 | 33.1ms |[下载链接](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.pdparams) |[下载链接](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.zip) | +| 目标跟踪(轻量级) | 视频输入 | MOTA: 69.1 | 27.2ms |[下载链接](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_s_36e_pipeline.pdparams) |[下载链接](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_s_36e_pipeline.zip) | | 属性识别 | 图片/视频输入 属性识别 | mA: 94.86 | 单人2ms | - |[下载链接](https://bj.bcebos.com/v1/paddledet/models/pipeline/strongbaseline_r50_30e_pa100k.zip) | | 关键点检测 | 视频输入 行为识别 | AP: 87.1 | 单人2.9ms |[下载链接](https://bj.bcebos.com/v1/paddledet/models/pipeline/dark_hrnet_w32_256x192.pdparams) |[下载链接](https://bj.bcebos.com/v1/paddledet/models/pipeline/dark_hrnet_w32_256x192.zip) | 行为识别 | 视频输入 行为识别 | 准确率: 96.43 | 单人2.7ms | - |[下载链接](https://bj.bcebos.com/v1/paddledet/models/pipeline/STGCN.zip) | diff --git a/deploy/pphuman/README_en.md b/deploy/pphuman/README_en.md index 735f817d4..07bff58b1 100644 --- a/deploy/pphuman/README_en.md +++ b/deploy/pphuman/README_en.md @@ -43,8 +43,10 @@ To make users have access to models of different scenarios, PP-Human provides pr | Task | Scenario | Precision | Inference Speed（FPS） | Model Weights |Model Inference and Deployment | | :---------: |:---------: |:--------------- | :-------: | :------: | :------: | -| Object Detection | Image/Video Input | mAP: 56.3 | 28.0ms |[Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.pdparams) |[Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.zip) | -| Object Tracking | Image/Video Input | MOTA: 72.0 | 33.1ms |[Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.pdparams) |[Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.zip) | +| Object Detection(high-precision) | Image/Video Input | mAP: 56.6 | 28.0ms |[Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.pdparams) |[Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.zip) | +| Object Detection(light-weight) | Image/Video Input | mAP: 53.2 | 22.1ms |[Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_s_36e_pipeline.pdparams) |[Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_s_36e_pipeline.zip) | +| Object Tracking(high-precision) | Image/Video Input | MOTA: 79.5 | 33.1ms |[Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.pdparams) |[Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.zip) | +| Object Tracking(light-weight) | Image/Video Input | MOTA: 69.1 | 27.2ms |[Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_s_36e_pipeline.pdparams) |[Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_s_36e_pipeline.zip) | | Attribute Recognition | Image/Video Input Attribute Recognition | mA: 94.86 | 2ms per person | - |[Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/strongbaseline_r50_30e_pa100k.zip) | | Keypoint Detection | Video Input Action Recognition | AP: 87.1 | 2.9ms per person | [Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/dark_hrnet_w32_256x192.pdparams) |[Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/dark_hrnet_w32_256x192.zip) | Action Recognition | Video Input Action Recognition | Precision 96.43 | 2.7ms per person | - |[Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/STGCN.zip) | diff --git a/deploy/pphuman/config/tracker_config.yml b/deploy/pphuman/config/tracker_config.yml index ddd55e865..50e92f7d7 100644 --- a/deploy/pphuman/config/tracker_config.yml +++ b/deploy/pphuman/config/tracker_config.yml @@ -2,7 +2,7 @@ # The tracker of MOT JDE Detector (such as FairMOT) is exported together with the model. # Here 'min_box_area' and 'vertical_ratio' are set for pedestrian, you can modify for other objects tracking. -type: JDETracker # 'JDETracker' or 'DeepSORTTracker' +type: JDETracker # BYTETracker JDETracker: @@ -13,14 +13,3 @@ JDETracker: match_thres: 0.9 min_box_area: 0 vertical_ratio: 0 # 1.6 for pedestrian - -DeepSORTTracker: - input_size: [64, 192] - min_box_area: 0 - vertical_ratio: -1 - budget: 100 - max_age: 70 - n_init: 3 - metric_type: cosine - matching_threshold: 0.2 - max_iou_distance: 0.9 diff --git a/deploy/pphuman/docs/mot.md b/deploy/pphuman/docs/mot.md index acfed453a..893ac479c 100644 --- a/deploy/pphuman/docs/mot.md +++ b/deploy/pphuman/docs/mot.md @@ -6,9 +6,10 @@ | 任务 | 算法 | 精度 | 预测速度(ms) |下载链接 | |:---------------------|:---------:|:------:|:------:| :---------------------------------------------------------------------------------: | -| 行人检测/跟踪 | PP-YOLOE | mAP: 56.3
MOTA: 72.0 | 检测: 28ms
跟踪：33.1ms | [下载链接](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.zip) | +| 行人检测/跟踪 | PP-YOLOE-l | mAP: 56.6
MOTA: 79.5 | 检测: 28.0ms
跟踪：33.1ms | [下载链接](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.zip) | +| 行人检测/跟踪 | PP-YOLOE-s | mAP: 53.2
MOTA: 69.1 | 检测: 22.1ms
跟踪：27.2ms | [下载链接](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_s_36e_pipeline.zip) | -1. 检测/跟踪模型精度为MOT17，CrowdHuman，HIEVE和部分业务数据融合训练测试得到 +1. 检测/跟踪模型精度为[COCO-Person](http://cocodataset.org/), [CrowdHuman](http://www.crowdhuman.org/), [HIEVE](http://humaninevents.org/) 和部分业务数据融合训练测试得到，验证集为业务数据 2. 预测速度为T4 机器上使用TensorRT FP16时的速度, 速度包含数据预处理、模型预测、后处理全流程 ## 使用方法 @@ -53,8 +54,8 @@ python deploy/pphuman/pipeline.py --config deploy/pphuman/config/infer_cfg.yml \ ## 方案说明 -1. 目标检测/多目标跟踪获取图片/视频输入中的行人检测框，模型方案为PP-YOLOE，详细文档参考[PP-YOLOE](../../../configs/ppyoloe/README_cn.md) -2. 多目标跟踪模型方案基于[ByteTrack](https://arxiv.org/pdf/2110.06864.pdf)，采用PP-YOLOE替换原文的YOLOX作为检测器，采用BYTETracker作为跟踪器。 +1. 目标检测/多目标跟踪获取图片/视频输入中的行人检测框，模型方案为PP-YOLOE，详细文档参考[PP-YOLOE](../../../configs/ppyoloe) +2. 多目标跟踪模型方案基于[ByteTrack](https://arxiv.org/pdf/2110.06864.pdf)，采用PP-YOLOE替换原文的YOLOX作为检测器，采用BYTETracker作为跟踪器，详细文档参考[ByteTrack](../../../configs/mot/bytetrack) ## 参考文献 ``` diff --git a/deploy/pphuman/docs/mot_en.md b/deploy/pphuman/docs/mot_en.md index 61a970898..510d86c06 100644 --- a/deploy/pphuman/docs/mot_en.md +++ b/deploy/pphuman/docs/mot_en.md @@ -6,9 +6,10 @@ Pedestrian detection and tracking is widely used in the intelligent community, i | Task | Algorithm | Precision | Inference Speed(ms) | Download Link | |:---------------------|:---------:|:------:|:------:| :---------------------------------------------------------------------------------: | -| Pedestrian Detection/ Tracking | PP-YOLOE | mAP: 56.3
MOTA: 72.0 | Detection: 28ms
Tracking：33.1ms | [Download Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.zip) | +| Pedestrian Detection/ Tracking | PP-YOLOE-l | mAP: 56.6
MOTA: 79.5 | Detection: 28.0ms
Tracking：33.1ms | [Download Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.zip) | +| Pedestrian Detection/ Tracking | PP-YOLOE-s | mAP: 53.2
MOTA: 69.2 | Detection: 22.1ms
Tracking：27.2ms | [Download Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_s_36e_pipeline.zip) | -1. The precision of the pedestrian detection/ tracking model is obtained by trainning and testing on [MOT17](https://motchallenge.net/), [CrowdHuman](http://www.crowdhuman.org/), [HIEVE](http://humaninevents.org/) and some business data. +1. The precision of the pedestrian detection/ tracking model is obtained by trainning and testing on [COCO-Person](http://cocodataset.org/), [CrowdHuman](http://www.crowdhuman.org/), [HIEVE](http://humaninevents.org/) and some business data. 2. The inference speed is the speed of using TensorRT FP16 on T4, the total number of data pre-training, model inference, and post-processing. ## How to Use @@ -57,7 +58,7 @@ Data source and copyright owner：Skyinfor Technology. Thanks for the provision 1. Get the pedestrian detection box of the image/ video input through object detection and multi-object tracking. The adopted model is PP-YOLOE, and for details, please refer to [PP-YOLOE](../../../configs/ppyoloe). -2. The multi-object tracking model solution is based on [ByteTrack](https://arxiv.org/pdf/2110.06864.pdf), and replace the original YOLOX with P-YOLOE as the detector，and BYTETracker as the tracker. +2. The multi-object tracking model solution is based on [ByteTrack](https://arxiv.org/pdf/2110.06864.pdf), and replace the original YOLOX with P-YOLOE as the detector，and BYTETracker as the tracker, please refer to [ByteTrack](../../../configs/mot/bytetrack). ## Reference ``` -- GitLab