English | 简体中文
- Release PP-YOLOE+ model: (2022.08)
- Pre training model using large-scale data set obj365
- In the backbone, add the alpha parameter to the block branch
- Optimize the end-to-end inference speed and improve the training convergence speed
- Please refer to:PP-YOLOE 2022.03 for details
PP-YOLOE is an excellent single-stage anchor-free model based on PP-YOLOv2, surpassing a variety of popular YOLO models. PP-YOLOE has a series of models, named s/m/l/x, which are configured through width multiplier and depth multiplier. PP-YOLOE avoids using special operators, such as Deformable Convolution or Matrix NMS, to be deployed friendly on various hardware. For more details, please refer to our report.
PP-YOLOE+_l achieves 53.3 mAP on COCO test-dev2017 dataset with 78.1 FPS on Tesla V100. While using TensorRT FP16, PP-YOLOE+_l can be further accelerated to 149.2 FPS. PP-YOLOE+_s/m/x also have excellent accuracy and speed performance, which can be found in Model Zoo
PP-YOLOE is composed of following methods:
- Scalable backbone and neck
- Task Alignment Learning
- Efficient Task-aligned head with DFL and VFL
- SiLU(Swish) activation function
Model | Epoch | GPU number | images/GPU | backbone | input shape | Box APval 0.5:0.95 |
Box APtest 0.5:0.95 |
Params(M) | FLOPs(G) | V100 FP32(FPS) | V100 TensorRT FP16(FPS) | download | config |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
PP-YOLOE+_s | 80 | 8 | 8 | cspresnet-s | 640 | 43.7 | 43.9 | 7.93 | 17.36 | 208.3 | 333.3 | model | config |
PP-YOLOE+_m | 80 | 8 | 8 | cspresnet-m | 640 | 49.8 | 50.0 | 23.43 | 49.91 | 123.4 | 208.3 | model | config |
PP-YOLOE+_l | 80 | 8 | 8 | cspresnet-l | 640 | 52.9 | 53.3 | 52.20 | 110.07 | 78.1 | 149.2 | model | config |
PP-YOLOE+_x | 80 | 8 | 8 | cspresnet-x | 640 | 54.7 | 54.9 | 98.42 | 206.59 | 45.0 | 95.2 | model | config |
Model | Epoch | AP0.5:0.95 | AP0.5 | AP0.75 | APsmall | APmedium | APlarge | ARsmall | ARmedium | ARlarge |
---|---|---|---|---|---|---|---|---|---|---|
PP-YOLOE+_s | 80 | 43.7 | 60.6 | 47.9 | 26.5 | 47.5 | 59.0 | 46.7 | 71.4 | 81.7 |
PP-YOLOE+_m | 80 | 49.8 | 67.1 | 54.5 | 31.8 | 53.9 | 66.2 | 53.3 | 75.0 | 84.6 |
PP-YOLOE+_l | 80 | 52.9 | 70.1 | 57.9 | 35.2 | 57.5 | 69.1 | 56.0 | 77.9 | 86.9 |
PP-YOLOE+_x | 80 | 54.7 | 72.0 | 59.9 | 37.9 | 59.3 | 70.4 | 57.0 | 78.7 | 87.2 |
Model | AP0.5:0.95 | TRT-FP32(fps) | TRT-FP16(fps) |
---|---|---|---|
PP-YOLOE+_s | 43.7 | 44.44 | 47.85 |
PP-YOLOE+_m | 49.8 | 39.06 | 43.86 |
PP-YOLOE+_l | 52.9 | 34.01 | 42.02 |
PP-YOLOE+_x | 54.7 | 26.88 | 36.76 |
Notes:
- PP-YOLOE is trained on COCO train2017 dataset and evaluated on val2017 & test-dev2017 dataset.
- The model weights in the table of Comprehensive Metrics are the same as that in the original Model Zoo, and evaluated on val2017.
- PP-YOLOE used 8 GPUs for mixed precision training, if GPU number or mini-batch size is changed, learning rate should be adjusted according to the formula lrnew = lrdefault * (batch_sizenew * GPU_numbernew) / (batch_sizedefault * GPU_numberdefault).
- PP-YOLOE inference speed is tesed on single Tesla V100 with batch size as 1, CUDA 10.2, CUDNN 7.6.5, TensorRT 6.0.1.8 in TensorRT mode.
- Refer to Speed testing to reproduce the speed testing results of PP-YOLOE.
- If you set
--run_benchmark=True
,you should install these dependencies at first,pip install pynvml psutil GPUtil
. - End-to-end speed test includes pre-processing + inference + post-processing and NMS time, using Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz, single Tesla V100, CUDA 11.2, CUDNN 8.2.0, TensorRT 8.0.1.6.
The PaddleDetection team provides configs and weights of various feature detection models based on PP-YOLOE, which users can download for use:
Scenarios | Related Datasets | Links |
---|---|---|
Pedestrian Detection | CrowdHuman | pphuman |
Vehicle Detection | BDD100K, UA-DETRAC | ppvehicle |
Small Object Detection | VisDrone | visdrone |
Training PP-YOLOE+ on 8 GPUs with following command
python -m paddle.distributed.launch --gpus 0,1,2,3,4,5,6,7 tools/train.py -c configs/ppyoloe/ppyoloe_plus_crn_l_80e_coco.yml --eval --amp
Notes:
- If you need to evaluate while training, please add
--eval
. - PP-YOLOE+ supports mixed precision training, please add
--amp
. - PaddleDetection supports multi-machine distribued training, you can refer to DistributedTraining tutorial.
Evaluating PP-YOLOE+ on COCO val2017 dataset in single GPU with following commands:
CUDA_VISIBLE_DEVICES=0 python tools/eval.py -c configs/ppyoloe/ppyoloe_plus_crn_l_80e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_l_80e_coco.pdparams
For evaluation on COCO test-dev2017 dataset, please download COCO test-dev2017 dataset from COCO dataset download and decompress to COCO dataset directory and configure EvalDataset
like configs/ppyolo/ppyolo_test.yml
.
Inference images in single GPU with following commands, use --infer_img
to inference a single image and --infer_dir
to inference all images in the directory.
# inference single image
CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c configs/ppyoloe/ppyoloe_plus_crn_l_80e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_l_80e_coco.pdparams --infer_img=demo/000000014439_640x640.jpg
# inference all images in the directory
CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c configs/ppyoloe/ppyoloe_plus_crn_l_80e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_l_80e_coco.pdparams --infer_dir=demo
For deployment on GPU or speed testing, model should be first exported to inference model using tools/export_model.py
.
Exporting PP-YOLOE+ for Paddle Inference without TensorRT, use following command
python tools/export_model.py -c configs/ppyoloe/ppyoloe_plus_crn_l_80e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_l_80e_coco.pdparams
Exporting PP-YOLOE+ for Paddle Inference with TensorRT for better performance, use following command with extra -o trt=True
setting.
python tools/export_model.py -c configs/ppyoloe/ppyoloe_plus_crn_l_80e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_l_80e_coco.pdparams trt=True
If you want to export PP-YOLOE model to ONNX format, use following command refer to PaddleDetection Model Export as ONNX Format Tutorial.
# export inference model
python tools/export_model.py -c configs/ppyoloe/ppyoloe_plus_crn_l_80e_coco.yml --output_dir=output_inference -o weights=https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_l_80e_coco.pdparams
# install paddle2onnx
pip install paddle2onnx
# convert to onnx
paddle2onnx --model_dir output_inference/ppyoloe_plus_crn_l_80e_coco --model_filename model.pdmodel --params_filename model.pdiparams --opset_version 11 --save_file ppyoloe_plus_crn_l_80e_coco.onnx
Notes: ONNX model only supports batch_size=1 now
For fair comparison, the speed in Model Zoo do not contains the time cost of data reading and post-processing(NMS), which is same as YOLOv4(AlexyAB) in testing method. Thus, you should export model with extra -o exclude_nms=True
setting.
Using Paddle Inference without TensorRT to test speed, run following command
# export inference model
python tools/export_model.py -c configs/ppyoloe/ppyoloe_plus_crn_l_80e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_l_80e_coco.pdparams exclude_nms=True
# speed testing with run_benchmark=True
CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/ppyoloe_plus_crn_l_80e_coco --image_file=demo/000000014439_640x640.jpg --run_mode=paddle --device=gpu --run_benchmark=True
Using Paddle Inference with TensorRT to test speed, run following command
# export inference model with trt=True
python tools/export_model.py -c configs/ppyoloe/ppyoloe_plus_crn_l_80e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_l_80e_coco.pdparams exclude_nms=True trt=True
# speed testing with run_benchmark=True,run_mode=trt_fp32/trt_fp16
CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/ppyoloe_plus_crn_l_80e_coco --image_file=demo/000000014439_640x640.jpg --run_mode=trt_fp16 --device=gpu --run_benchmark=True
Using TensorRT Inference with ONNX to test speed, run following command
# export inference model with trt=True
python tools/export_model.py -c configs/ppyoloe/ppyoloe_plus_crn_s_80e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_s_80e_coco.pdparams exclude_nms=True trt=True
# convert to onnx
paddle2onnx --model_dir output_inference/ppyoloe_plus_crn_s_80e_coco --model_filename model.pdmodel --params_filename model.pdiparams --opset_version 12 --save_file ppyoloe_plus_crn_s_80e_coco.onnx
# trt inference using fp16 and batch_size=1
trtexec --onnx=./ppyoloe_plus_crn_s_80e_coco.onnx --saveEngine=./ppyoloe_s_bs1.engine --workspace=1024 --avgRuns=1000 --shapes=image:1x3x640x640,scale_factor:1x2 --fp16
# trt inference using fp16 and batch_size=32
trtexec --onnx=./ppyoloe_plus_crn_s_80e_coco.onnx --saveEngine=./ppyoloe_s_bs32.engine --workspace=1024 --avgRuns=1000 --shapes=image:32x3x640x640,scale_factor:32x2 --fp16
# Using the above script, T4 and tensorrt 7.2 machine, the speed of PPYOLOE-s model is as follows,
# batch_size=1, 2.80ms, 357fps
# batch_size=32, 67.69ms, 472fps
PP-YOLOE can be deployed by following approches:
- Paddle Inference Python & C++
- Paddle-TensorRT
- PaddleServing
- PaddleSlim
Next, we will introduce how to use Paddle Inference to deploy PP-YOLOE models in TensorRT FP16 mode.
First, refer to Paddle Inference Docs, download and install packages corresponding to CUDA, CUDNN and TensorRT version.
Then, Exporting PP-YOLOE for Paddle Inference with TensorRT, use following command.
python tools/export_model.py -c configs/ppyoloe/ppyoloe_plus_crn_l_80e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_l_80e_coco.pdparams trt=True
Finally, inference in TensorRT FP16 mode.
# inference single image
CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/ppyoloe_plus_crn_l_80e_coco --image_file=demo/000000014439_640x640.jpg --device=gpu --run_mode=trt_fp16
# inference all images in the directory
CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/ppyoloe_plus_crn_l_80e_coco --image_dir=demo/ --device=gpu --run_mode=trt_fp16
Notes:
- TensorRT will perform optimization for the current hardware platform according to the definition of the network, generate an inference engine and serialize it into a file. This inference engine is only applicable to the current hardware hardware platform. If your hardware and software platform has not changed, you can set
use_static=True
in enable_tensorrt_engine. In this way, the serialized file generated will be saved in theoutput_inference
folder, and the saved serialized file will be loaded the next time when TensorRT is executed. - PaddleDetection release/2.4 and later versions will support NMS calling TensorRT, which requires PaddlePaddle release/2.3 and later versions.
Model | AP | AP50 |
---|---|---|
YOLOX | 22.6 | 37.5 |
YOLOv5 | 26.0 | 42.7 |
PP-YOLOE | 30.5 | 46.4 |
Notes
- Here, we use VisDrone dataset, and to detect 9 objects including
person, bicycles, car, van, truck, tricyle, awning-tricyle, bus, motor
. - Above models trained using official default config, and load pretrained parameters on COCO dataset.
- Due to the limited time, more verification results will be supplemented in the future. You are also welcome to contribute to PP-YOLOE
Ablation experiments of PP-YOLOE.
NO. | Model | Box APval | Params(M) | FLOPs(G) | V100 FP32 FPS |
---|---|---|---|---|---|
A | PP-YOLOv2 | 49.1 | 54.58 | 115.77 | 68.9 |
B | A + Anchor-free | 48.8 | 54.27 | 114.78 | 69.8 |
C | B + CSPRepResNet | 49.5 | 47.42 | 101.87 | 85.5 |
D | C + TAL | 50.4 | 48.32 | 104.75 | 84.0 |
E | D + ET-Head | 50.9 | 52.20 | 110.07 | 78.1 |