PaddleDetection yolov5

https://github.com/Sharpiless/PaddleDetection-Yolov5

简介

PaddleDetection飞桨目标检测开发套件，旨在帮助开发者更快更好地完成检测模型的组建、训练、优化及部署等全开发流程。

PaddleDetection模块化地实现了多种主流目标检测算法，提供了丰富的数据增强策略、网络模块组件（如骨干网络）、损失函数等，并集成了模型压缩和跨平台高性能部署能力。

经过长时间产业实践打磨，PaddleDetection已拥有顺畅、卓越的使用体验，被工业质检、遥感图像检测、无人巡检、新零售、互联网、科研等十多个行业的开发者广泛应用。

Yolov5：

YOLOV4出现之后不久，YOLOv5横空出世。YOLOv5在YOLOv4算法的基础上做了进一步的改进，检测性能得到进一步的提升。虽然YOLOv5算法并没有与YOLOv4算法进行性能比较与分析，但是YOLOv5在COCO数据集上面的测试效果还是挺不错的。大家对YOLOv5算法的创新性半信半疑，有的人对其持肯定态度，有的人对其持否定态度。在我看来，YOLOv5检测算法中还是存在很多可以学习的地方，虽然这些改进思路看来比较简单或者创新点不足，但是它们确定可以提升检测算法的性能。其实工业界往往更喜欢使用这些方法，而不是利用一个超级复杂的算法来获得较高的检测精度。本文将对YOLOv5检测算法进行复现。

下载预训练模型：

https://drive.google.com/file/d/16tREOOJzKgOLw31bSiUNz0iBdqoRzq1i/view?usp=sharing

训练Yolov5：

python tools/train.py -c configs/yolov5/yolov5s_CSPdarknet_roadsign.yml

实验结果：

0.9087 mAP on roadsign dataset.

关注我的公众号：

感兴趣的同学关注我的公众号——可达鸭的深度学习教程：

联系作者：

B站：https://space.bilibili.com/470550823

CSDN：https://blog.csdn.net/weixin_44936889

AI Studio：https://aistudio.baidu.com/aistudio/personalcenter/thirdview/67156

Github：https://github.com/Sharpiless

%cd work/

/home/aistudio/work

!unzip PPDet-yolov5v2.zip -d ./

!python tools/train.py -c configs/yolov5/yolov5s_CSPdarknet_roadsign.yml --eval

/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/tensor/creation.py:125: DeprecationWarning: `np.object` is a deprecated alias for the builtin `object`. To silence this warning, use `object` by itself. Doing this will not modify any behavior and is safe. 
Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations
  if data.dtype == np.object:
[07/15 10:17:41] ppdet.utils.download WARNING: Config annotation dataset/roadsign_voc/train.txt is not a file, dataset config is not valid
[07/15 10:17:41] ppdet.utils.download INFO: Dataset /home/aistudio/work/dataset/roadsign_voc is not valid for reason above, try searching /home/aistudio/.cache/paddle/dataset or downloading dataset...
[07/15 10:17:41] ppdet.utils.download INFO: Found /home/aistudio/.cache/paddle/dataset/roadsign_voc/annotations
[07/15 10:17:41] ppdet.utils.download INFO: Found /home/aistudio/.cache/paddle/dataset/roadsign_voc/images
[07/15 10:17:41] reader WARNING: Shared memory size is less than 1G, disable shared_memory in DataLoader
[07/15 10:17:42] ppdet.utils.checkpoint INFO: Finish loading model weights: output.pdparams
[07/15 10:17:51] ppdet.engine INFO: Epoch: [0] [ 0/87] learning_rate: 0.000033 loss_xy: 0.752040 loss_wh: 0.698217 loss_iou: 2.634957 loss_obj: 11.301561 loss_cls: 1.041652 loss: 16.428429 eta: 8:28:32 batch_cost: 8.7679 data_cost: 0.9061 ips: 0.9124 images/s
[07/15 10:19:42] ppdet.engine INFO: Epoch: [0] [20/87] learning_rate: 0.000047 loss_xy: 0.529626 loss_wh: 0.569290 loss_iou: 2.436198 loss_obj: 8.576855 loss_cls: 1.023474 loss: 13.317031 eta: 5:29:28 batch_cost: 5.5608 data_cost: 0.0002 ips: 1.4386 images/s
[07/15 10:21:42] ppdet.engine INFO: Epoch: [0] [40/87] learning_rate: 0.000060 loss_xy: 0.500230 loss_wh: 0.502719 loss_iou: 2.226187 loss_obj: 4.208471 loss_cls: 0.890207 loss: 8.235611 eta: 5:35:40 batch_cost: 6.0032 data_cost: 0.0003 ips: 1.3326 images/s
[07/15 10:23:23] ppdet.engine INFO: Epoch: [0] [60/87] learning_rate: 0.000073 loss_xy: 0.519860 loss_wh: 0.599364 loss_iou: 2.455585 loss_obj: 3.626266 loss_cls: 1.031202 loss: 8.345335 eta: 5:18:38 batch_cost: 5.0474 data_cost: 0.0003 ips: 1.5850 images/s
[07/15 10:25:13] ppdet.engine INFO: Epoch: [0] [80/87] learning_rate: 0.000087 loss_xy: 0.568008 loss_wh: 0.618775 loss_iou: 2.583227 loss_obj: 3.632595 loss_cls: 0.863238 loss: 7.575019 eta: 5:15:29 batch_cost: 5.4984 data_cost: 0.0002 ips: 1.4550 images/s
[07/15 10:25:47] ppdet.utils.checkpoint INFO: Save checkpoint: output/yolov5s_CSPdarknet_roadsign
[07/15 10:25:47] ppdet.utils.download WARNING: Config annotation dataset/roadsign_voc/valid.txt is not a file, dataset config is not valid
[07/15 10:25:47] ppdet.utils.download INFO: Dataset /home/aistudio/work/dataset/roadsign_voc is not valid for reason above, try searching /home/aistudio/.cache/paddle/dataset or downloading dataset...
[07/15 10:25:47] ppdet.utils.download INFO: Found /home/aistudio/.cache/paddle/dataset/roadsign_voc/annotations
[07/15 10:25:47] ppdet.utils.download INFO: Found /home/aistudio/.cache/paddle/dataset/roadsign_voc/images
[07/15 10:25:48] ppdet.engine INFO: Eval iter: 0
[07/15 10:26:09] ppdet.engine INFO: Eval iter: 100
[07/15 10:26:25] ppdet.metrics.metrics INFO: Accumulating evaluatation results...
[07/15 10:26:25] ppdet.metrics.metrics INFO: mAP(0.50, integral) = 85.84%
[07/15 10:26:25] ppdet.engine INFO: Total sample number: 176, averge FPS: 4.751870228058035
[07/15 10:26:25] ppdet.engine INFO: Best test bbox ap is 0.858.
[07/15 10:26:25] ppdet.utils.checkpoint INFO: Save checkpoint: output/yolov5s_CSPdarknet_roadsign
[07/15 10:26:35] ppdet.engine INFO: Epoch: [1] [ 0/87] learning_rate: 0.000091 loss_xy: 0.567437 loss_wh: 0.623783 loss_iou: 2.511684 loss_obj: 3.314124 loss_cls: 0.949793 loss: 7.338743 eta: 5:16:15 batch_cost: 6.2481 data_cost: 0.0003 ips: 1.2804 images/s
[07/15 10:28:39] ppdet.engine INFO: Epoch: [1] [20/87] learning_rate: 0.000100 loss_xy: 0.583728 loss_wh: 0.708465 loss_iou: 2.704193 loss_obj: 3.461134 loss_cls: 1.127932 loss: 9.057523 eta: 5:20:59 batch_cost: 6.2270 data_cost: 0.0003 ips: 1.2847 images/s
[07/15 10:30:28] ppdet.engine INFO: Epoch: [1] [40/87] learning_rate: 0.000100 loss_xy: 0.576615 loss_wh: 0.655194 loss_iou: 2.566234 loss_obj: 2.921384 loss_cls: 1.010778 loss: 7.844104 eta: 5:16:43 batch_cost: 5.4392 data_cost: 0.0003 ips: 1.4708 images/s
[07/15 10:32:34] ppdet.engine INFO: Epoch: [1] [60/87] learning_rate: 0.000100 loss_xy: 0.583071 loss_wh: 0.726098 loss_iou: 2.730413 loss_obj: 3.053501 loss_cls: 0.991524 loss: 8.496977 eta: 5:19:40 batch_cost: 6.3128 data_cost: 0.0003 ips: 1.2673 images/s
[07/15 10:34:31] ppdet.engine INFO: Epoch: [1] [80/87] learning_rate: 0.000100 loss_xy: 0.606061 loss_wh: 0.652358 loss_iou: 2.841094 loss_obj: 3.237591 loss_cls: 1.084277 loss: 8.605825 eta: 5:18:16 batch_cost: 5.8318 data_cost: 0.0003 ips: 1.3718 images/s
[07/15 10:34:59] ppdet.utils.checkpoint INFO: Save checkpoint: output/yolov5s_CSPdarknet_roadsign
[07/15 10:35:00] ppdet.engine INFO: Eval iter: 0
[07/15 10:35:19] ppdet.engine INFO: Eval iter: 100
[07/15 10:35:33] ppdet.metrics.metrics INFO: Accumulating evaluatation results...
[07/15 10:35:33] ppdet.metrics.metrics INFO: mAP(0.50, integral) = 85.30%
[07/15 10:35:33] ppdet.engine INFO: Total sample number: 176, averge FPS: 5.151774310709877
[07/15 10:35:33] ppdet.engine INFO: Best test bbox ap is 0.858.
[07/15 10:35:46] ppdet.engine INFO: Epoch: [2] [ 0/87] learning_rate: 0.000100 loss_xy: 0.537015 loss_wh: 0.587401 loss_iou: 2.352699 loss_obj: 3.121367 loss_cls: 1.012583 loss: 7.857001 eta: 5:17:11 batch_cost: 5.8271 data_cost: 0.0003 ips: 1.3729 images/s
^C

!rm -rf output/

!zip -r code.zip ./*

基于Paddlepaddle复现yolov5，支持PaddleDetection接口

Related tags

Overview

PaddleDetection yolov5

简介

Yolov5：

下载预训练模型：

训练Yolov5：

实验结果：

关注我的公众号：

联系作者：

Owner

Official PyTorch implementation of "IntegralAction: Pose-driven Feature Integration for Robust Human Action Recognition in Videos", CVPRW 2021

For visualizing the dair-v2x-i dataset

CRISCE: Automatically Generating Critical Driving Scenarios From Car Accident Sketches

CRNN With PyTorch

Github Traffic Insights as Prometheus metrics.

A set of examples around hub for creating and processing datasets

Script for getting information in discord

Code for 'Blockwise Sequential Model Learning for Partially Observable Reinforcement Learning' (AAAI 2022)

Pretrained models for Jax/Haiku; MobileNet, ResNet, VGG, Xception.

CPF: Learning a Contact Potential Field to Model the Hand-object Interaction

Learning to Initialize Neural Networks for Stable and Efficient Training

A criticism of a recent paper on buggy image downsampling methods in popular image processing and deep learning libraries.

Datasets, Transforms and Models specific to Computer Vision

The source code of the ICCV2021 paper "PIRenderer: Controllable Portrait Image Generation via Semantic Neural Rendering"

This MVP data web app uses the Streamlit framework and Facebook's Prophet forecasting package to generate a dynamic forecast from your own data.

A Learning-based Camera Calibration Toolbox

TensorFlow implementation of PHM (Parameterization of Hypercomplex Multiplication)

JASS: Japanese-specific Sequence to Sequence Pre-training for Neural Machine Translation

RIFE - Real-Time Intermediate Flow Estimation for Video Frame Interpolation

Learning based AI for playing multi-round Koi-Koi hanafuda card games. Have fun.