diff --git a/configs/datasets/coco_detection.yml b/configs/datasets/coco_detection.yml index 7a62c3b0b57a5d76c8ed519d3a3940c1b4532c15..614135743712b195a8d3822609d9a3c28ee60d22 100644 --- a/configs/datasets/coco_detection.yml +++ b/configs/datasets/coco_detection.yml @@ -17,3 +17,4 @@ EvalDataset: TestDataset: !ImageFolder anno_path: annotations/instances_val2017.json + dataset_dir: dataset/coco diff --git a/configs/datasets/coco_instance.yml b/configs/datasets/coco_instance.yml index 5eaf76791a94bfd2819ba6dab610fae54b69f26e..5b074b00d73ced75247b6ad56604e7c90b705c1a 100644 --- a/configs/datasets/coco_instance.yml +++ b/configs/datasets/coco_instance.yml @@ -17,3 +17,4 @@ EvalDataset: TestDataset: !ImageFolder anno_path: annotations/instances_val2017.json + dataset_dir: dataset/coco diff --git a/configs/datasets/dota.yml b/configs/datasets/dota.yml index f9d9395b00d7ed9028396044c407784d251e43e5..5153163d95a8a418a82d3d6d43f6e1f9404ed075 100644 --- a/configs/datasets/dota.yml +++ b/configs/datasets/dota.yml @@ -17,3 +17,4 @@ EvalDataset: TestDataset: !ImageFolder anno_path: trainval_split/s2anet_trainval_paddle_coco.json + dataset_dir: dataset/DOTA_1024_s2anet/ diff --git a/configs/mot/bytetrack/_base_/mot17.yml b/configs/mot/bytetrack/_base_/mot17.yml index 2efa55546026168c39396c4d51a71428e19a0638..faf47f622d1c2847a9686dfa8d7e48a49c05436c 100644 --- a/configs/mot/bytetrack/_base_/mot17.yml +++ b/configs/mot/bytetrack/_base_/mot17.yml @@ -17,6 +17,7 @@ EvalDataset: TestDataset: !ImageFolder + dataset_dir: dataset/mot/MOT17 anno_path: annotations/val_half.json diff --git a/configs/mot/deepsort/_base_/mot17.yml b/configs/mot/deepsort/_base_/mot17.yml index 2efa55546026168c39396c4d51a71428e19a0638..faf47f622d1c2847a9686dfa8d7e48a49c05436c 100644 --- a/configs/mot/deepsort/_base_/mot17.yml +++ b/configs/mot/deepsort/_base_/mot17.yml @@ -17,6 +17,7 @@ EvalDataset: TestDataset: !ImageFolder + dataset_dir: dataset/mot/MOT17 anno_path: annotations/val_half.json diff --git a/deploy/pphuman/README.md b/deploy/pphuman/README.md index 008cb69b170b5e60dbf071c9ad0c6ca9a466e4b8..af61b3a253670dbe3a03b7b2ebeb54b5b9668946 100644 --- a/deploy/pphuman/README.md +++ b/deploy/pphuman/README.md @@ -92,20 +92,26 @@ ATTR: ### 3. 预测部署 ``` -# 指定配置文件路径和测试图片 +# 行人检测，指定配置文件路径和测试图片 python deploy/pphuman/pipeline.py --config deploy/pphuman/config/infer_cfg.yml --image_file=test_image.jpg --device=gpu -# 指定配置文件路径和测试视频，完成属性识别 +# 行人跟踪，指定配置文件路径和测试视频 +python deploy/pphuman/pipeline.py --config deploy/pphuman/config/infer_cfg.yml --video_file=test_video.mp4 --device=gpu + +# 行人跟踪，指定配置文件路径，模型路径和测试视频 +# 命令行中指定的模型路径优先级高于配置文件 +python deploy/pphuman/pipeline.py --config deploy/pphuman/config/infer_cfg.yml --video_file=test_video.mp4 --device=gpu --model_dir det=ppyoloe/ + +# 行人属性识别，指定配置文件路径和测试视频 python deploy/pphuman/pipeline.py --config deploy/pphuman/config/infer_cfg.yml --video_file=test_video.mp4 --device=gpu --enable_attr=True -# 指定配置文件路径和测试视频，完成行为识别 +# 行为识别，指定配置文件路径和测试视频 python deploy/pphuman/pipeline.py --config deploy/pphuman/config/infer_cfg.yml --video_file=test_video.mp4 --device=gpu --enable_action=True -# 指定配置文件路径，模型路径和测试视频，完成多目标跟踪 -# 命令行中指定的模型路径优先级高于配置文件 -python deploy/pphuman/pipeline.py --config deploy/pphuman/config/infer_cfg.yml --video_file=test_video.mp4 --device=gpu --model_dir det=ppyoloe/ ``` +其他用法请参考[子任务文档](./docs) + #### 3.1 参数说明 | 参数 | 是否必须|含义 | @@ -136,17 +142,17 @@ PP-Human整体方案如下图所示 -### 1. 目标检测 +### 1. 行人检测 - 采用PP-YOLOE L 作为目标检测模型 - 详细文档参考[PP-YOLOE](../../configs/ppyoloe/)和[检测跟踪文档](docs/mot.md) -### 2. 多目标跟踪 -- 采用SDE方案完成多目标跟踪 +### 2. 行人跟踪 +- 采用SDE方案完成行人跟踪 - 检测模型使用PP-YOLOE L - 跟踪模块采用Bytetrack方案 - 详细文档参考[Bytetrack](../../configs/mot/bytetrack)和[检测跟踪文档](docs/mot.md) -### 3. 跨镜跟踪 +### 3. 跨镜行人跟踪 - 使用PP-YOLOE + Bytetrack得到单镜头多目标跟踪轨迹 - 使用ReID（centroid网络）对每一帧的检测结果提取特征 - 多镜头轨迹特征进行匹配，得到跨镜头跟踪结果 diff --git a/deploy/pphuman/README_en.md b/deploy/pphuman/README_en.md index d45481b4d5f8da088b251997bbe8266146b72c9f..44c6f78632ab6966b6c38f003e0657e499f87c87 100644 --- a/deploy/pphuman/README_en.md +++ b/deploy/pphuman/README_en.md @@ -3,11 +3,11 @@ English | [简体中文](README.md) # PP-Human— a Real-Time Pedestrian Analysis Tool PP-Human serves as the first open-source tool of real-time pedestrian anaylsis relying on the PaddlePaddle deep learning framework. Versatile and efficient in deployment, it has been used in various senarios. PP-Human -offers many input options, including image/single-camera video/multi-camera video, and covers multi-object tracking, attribute recognition, and action recognition. PP-Human can be applied to intelligent traffic, the intelligent community, industiral patrol, and so on. It supports server-side deployment and TensorRT acceleration，and even can achieve real-time analysis on the T4 server. +offers many input options, including image/single-camera video/multi-camera video, and covers multi-object tracking, attribute recognition, and action recognition. PP-Human can be applied to intelligent traffic, the intelligent community, industiral patrol, and so on. It supports server-side deployment and TensorRT acceleration，and achieves real-time analysis on the T4 server. ## I. Environment Preparation -Requirement: PaddleDetection version >= release/2.4 +Requirement: PaddleDetection version >= release/2.4 or develop The installation of PaddlePaddle and PaddleDetection @@ -38,18 +38,19 @@ To make users have access to models of different scenarios, PP-Human provides pr | Task | Scenario | Precision | Inference Speed（FPS） | Model Inference and Deployment | | :---------: |:---------: |:--------------- | :-------: | :------: | -| Object Detection | Image/Video Input | - | - | [Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.zip) | -| Attribute Recognition | Image/Video Input Attribute Recognition | - | - | [Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/strongbaseline_r50_30e_pa100k.tar) | -| Keypoint Detection | Video Input Action Recognition | - | - | [Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/dark_hrnet_w32_256x192.zip) -| Behavior Recognition | Video Input Bheavior Recognition | - | - | [Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/STGCN.zip) | -| ReID | Multi-Target Multi-Camera Tracking | - | - | [Link]() | +| Object Detection | Image/Video Input | mAP: 56.3 | 28.0ms | [Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.zip) | +| Attribute Recognition | Image/Video Input Attribute Recognition | MOTA: 72.0 | 33.1ms | [Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/strongbaseline_r50_30e_pa100k.zip) | +| Keypoint Detection | Video Input Action Recognition | mA: 94.86 | 2ms per person | [Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/dark_hrnet_w32_256x192.zip) +| Behavior Recognition | Video Input Bheavior Recognition | Precision 96.43 | 2.7ms per person | [Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/STGCN.zip) | +| ReID | Multi-Target Multi-Camera Tracking | mAP: 99.7 | - | [Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/reid_model.zip) | Then, unzip the downloaded model to the folder `./output_inference`. **Note: ** - The model precision is decided by the fusion of datasets which include open-source datasets and enterprise ones. -- When the inference speed is T4, use TensorRT FP16. +- The precision on ReID model is evaluated on Market1501. +- The inference speed is tested on T4, using TensorRT FP16. The pipeline of preprocess, prediction and postprocess is included. ### 2. Preparation of Configuration Files @@ -80,31 +81,44 @@ ATTR: batch_size: 8 ``` +**Note: ** + +- For different tasks, users could add `--enable_attr=True` or `--enable_action=True` in command line and do not need to set config file. +- if only need to change the model path, users could add `--model_dir det=ppyoloe/` in command line and do not need to set config file. For details info please refer to doc below. ### 3. Inference and Deployment ``` -# Specify the config file path and test images +# Pedestrian detection. Specify the config file path and test images python deploy/pphuman/pipeline.py --config deploy/pphuman/config/infer_cfg.yml --image_file=test_image.jpg --device=gpu -# Specify the config file path and test videos，and finish the attribute recognition +# Pedestrian tracking. Specify the config file path and test videos +python deploy/pphuman/pipeline.py --config deploy/pphuman/config/infer_cfg.yml --video_file=test_video.mp4 --device=gpu + +# Pedestrian tracking. Specify the config file path, the model path and test videos +# The model path specified on the command line prioritizes over the config file +python deploy/pphuman/pipeline.py --config deploy/pphuman/config/infer_cfg.yml --video_file=test_video.mp4 --device=gpu --model_dir det=ppyoloe/ + +# Attribute recognition. Specify the config file path and test videos python deploy/pphuman/pipeline.py --config deploy/pphuman/config/infer_cfg.yml --video_file=test_video.mp4 --device=gpu --enable_attr=True -# Specify the config file path and test videos，and finish the Action Recognition +# Action Recognition. Specify the config file path and test videos python deploy/pphuman/pipeline.py --config deploy/pphuman/config/infer_cfg.yml --video_file=test_video.mp4 --device=gpu --enable_action=True -# Specify the config file path, the model path and test videos，and finish the multi-object tracking -# The model path specified on the command line prioritizes over the config file -python deploy/pphuman/pipeline.py --config deploy/pphuman/config/infer_cfg.yml --video_file=test_video.mp4 --device=gpu --model_dir det=ppyoloe/ +# Multi-Camera pedestrian tracking. Specify the config file path and test videos +python deploy/pphuman/pipeline.py --config deploy/pphuman/config/infer_cfg.yml --video_dir=test_video_dir/ --device=gpu + ``` +Other usage please refer to [sub-task docs](./docs) + ### 3.1 Description of Parameters | Parameter | Optional or not| Meaning | |-------|-------|----------| | --config | Yes | Config file path | -| --model_dir | Option | the model paths of different tasks in PP-Human, with a priority higher than config files | +| --model_dir | Option | the model paths of different tasks in PP-Human, with a priority higher than config files. For example, `--model_dir det=better_det/ attr=better_attr/` | | --image_file | Option | Images to-be-predicted | | --image_dir | Option | The path of folders of to-be-predicted images | | --video_file | Option | Videos to-be-predicted | @@ -130,21 +144,21 @@ The overall solution of PP-Human is as follows: ### 1. Object Detection - Use PP-YOLOE L as the model of object detection -- For details, please refer to [PP-YOLOE](../../configs/ppyoloe/) +- For details, please refer to [PP-YOLOE](../../configs/ppyoloe/) and [Detection and Tracking](docs/mot_en.md) ### 2. Multi-Object Tracking - Conduct multi-object tracking with the SDE solution - Use PP-YOLOE L as the detection model - Use the Bytetrack solution to track modules -- For details, refer to [Bytetrack](configs/mot/bytetrack) +- For details, refer to [Bytetrack](configs/mot/bytetrack) and [Detection and Tracking](docs/mot_en.md) -### 3. Cross-Camera Tracking +### 3. Multi-Camera Tracking - Use PP-YOLOE + Bytetrack to obtain the tracks of single-camera multi-object tracking - Use ReID（centroid network）to extract features of the detection result of each frame - Match the features of multi-camera tracks to get the cross-camera tracking result -- For details, please refer to [Cross-Camera Tracking](docs/mtmct_en.md) +- For details, please refer to [Multi-Camera Tracking](docs/mtmct_en.md) -### 4. Multi-Target Multi-Camera Tracking +### 4. Attribute Recognition - Use PP-YOLOE + Bytetrack to track humans - Use StrongBaseline（a multi-class model）to conduct attribute recognition, and the main attributes include age, gender, hats, eyes, clothing, and backpacks. - For details, please refer to [Attribute Recognition](docs/attribute_en.md) diff --git a/ppdet/data/source/category.py b/ppdet/data/source/category.py index 9390e54c4ce5dacce4674363689b629261c787c6..2c366968a439088bbec66ad2fc5ac182daea199c 100644 --- a/ppdet/data/source/category.py +++ b/ppdet/data/source/category.py @@ -39,6 +39,11 @@ def get_categories(metric_type, anno_file=None, arch=None): if arch == 'keypoint_arch': return (None, {'id': 'keypoint'}) + if anno_file == None or (not os.path.isfile(anno_file)): + logger.warning("anno_file '{}' is None or not set or not exist, " + "please recheck TrainDataset/EvalDataset/TestDataset.anno_path, " + "otherwise the default categories will be used by metric_type.".format(anno_file)) + if metric_type.lower() == 'coco' or metric_type.lower( ) == 'rbox' or metric_type.lower() == 'snipercoco': if anno_file and os.path.isfile(anno_file): @@ -55,8 +60,9 @@ def get_categories(metric_type, anno_file=None, arch=None): # anno file not exist, load default categories of COCO17 else: if metric_type.lower() == 'rbox': + logger.warning("metric_type: {}, load default categories of DOTA.".format(metric_type)) return _dota_category() - + logger.warning("metric_type: {}, load default categories of COCO.".format(metric_type)) return _coco17_category() elif metric_type.lower() == 'voc': @@ -77,6 +83,7 @@ def get_categories(metric_type, anno_file=None, arch=None): # anno file not exist, load default categories of # VOC all 20 categories else: + logger.warning("metric_type: {}, load default categories of VOC.".format(metric_type)) return _vocall_category() elif metric_type.lower() == 'oid': @@ -104,6 +111,7 @@ def get_categories(metric_type, anno_file=None, arch=None): return clsid2catid, catid2name # anno file not exist, load default category 'pedestrian'. else: + logger.warning("metric_type: {}, load default categories of pedestrian MOT.".format(metric_type)) return _mot_category(category='pedestrian') elif metric_type.lower() in ['kitti', 'bdd100kmot']: @@ -122,6 +130,7 @@ def get_categories(metric_type, anno_file=None, arch=None): return clsid2catid, catid2name # anno file not exist, load default categories of visdrone all 10 categories else: + logger.warning("metric_type: {}, load default categories of VisDrone.".format(metric_type)) return _visdrone_category() else: diff --git a/ppdet/metrics/mcmot_metrics.py b/ppdet/metrics/mcmot_metrics.py index 5bcfb923470a1f94a2bd951fb721221a8f339354..48d15c90e6512eb8943ca3ee224ac92b2795453c 100644 --- a/ppdet/metrics/mcmot_metrics.py +++ b/ppdet/metrics/mcmot_metrics.py @@ -26,8 +26,6 @@ from motmetrics.math_util import quiet_divide import numpy as np import pandas as pd -import paddle -import paddle.nn.functional as F from .metrics import Metric import motmetrics as mm import openpyxl @@ -311,7 +309,9 @@ class MCMOTEvaluator(object): self.gt_filename = os.path.join(self.data_root, '../', 'sequences', '{}.txt'.format(self.seq_name)) - + if not os.path.exists(self.gt_filename): + logger.warning("gt_filename '{}' of MCMOTEvaluator is not exist, so the MOTA will be -inf.") + def reset_accumulator(self): import motmetrics as mm mm.lap.default_solver = 'lap' diff --git a/ppdet/metrics/mot_metrics.py b/ppdet/metrics/mot_metrics.py index 85cba3630cd428478175ddc6347db4152d47a353..af2f7dd19c801cfe2c34d86c6c77ed4816b6fbec 100644 --- a/ppdet/metrics/mot_metrics.py +++ b/ppdet/metrics/mot_metrics.py @@ -22,8 +22,7 @@ import sys import math from collections import defaultdict import numpy as np -import paddle -import paddle.nn.functional as F + from ppdet.modeling.bbox_utils import bbox_iou_np_expand from .map_utils import ap_per_class from .metrics import Metric @@ -36,8 +35,10 @@ __all__ = ['MOTEvaluator', 'MOTMetric', 'JDEDetMetric', 'KITTIMOTMetric'] def read_mot_results(filename, is_gt=False, is_ignore=False): - valid_labels = {1} - ignore_labels = {2, 7, 8, 12} # only in motchallenge datasets like 'MOT16' + valid_label = [1] + ignore_labels = [2, 7, 8, 12] # only in motchallenge datasets like 'MOT16' + logger.info("In MOT16/17 dataset the valid_label of ground truth is '{}', " + "in other dataset it should be '0' for single classs MOT.".format(valid_label[0])) results_dict = dict() if os.path.isfile(filename): with open(filename, 'r') as f: @@ -50,12 +51,10 @@ def read_mot_results(filename, is_gt=False, is_ignore=False): continue results_dict.setdefault(fid, list()) - box_size = float(linelist[4]) * float(linelist[5]) - if is_gt: label = int(float(linelist[7])) mark = int(float(linelist[6])) - if mark == 0 or label not in valid_labels: + if mark == 0 or label not in valid_label: continue score = 1 elif is_ignore: @@ -118,6 +117,8 @@ class MOTEvaluator(object): assert self.data_type == 'mot' gt_filename = os.path.join(self.data_root, self.seq_name, 'gt', 'gt.txt') + if not os.path.exists(gt_filename): + logger.warning("gt_filename '{}' of MOTEvaluator is not exist, so the MOTA will be -inf.") self.gt_frame_dict = read_mot_results(gt_filename, is_gt=True) self.gt_ignore_frame_dict = read_mot_results( gt_filename, is_ignore=True) diff --git a/ppdet/modeling/tests/test_architectures.py b/ppdet/modeling/tests/test_architectures.py index 25767e74abd9fce29c51adbc4b5109e17a50aa0b..5de79b2cedb3fffac0ce853406560821a9142363 100644 --- a/ppdet/modeling/tests/test_architectures.py +++ b/ppdet/modeling/tests/test_architectures.py @@ -62,7 +62,7 @@ class TestGFL(TestFasterRCNN): class TestPicoDet(TestFasterRCNN): def set_config(self): - self.cfg_file = 'configs/picodet/picodet_s_320_coco.yml' + self.cfg_file = 'configs/picodet/picodet_s_320_coco_lcnet.yml' if __name__ == '__main__':