基于目标检测的番茄采摘模型

转自AI Studio，原文链接：【AI Workshop】PaddleDetection番茄目标果检测 - 飞桨AI Studio1、项目内容随着日常饮食的逐渐丰富,以番茄为原料的食品种类也在不断地增多。在当下劳动力日渐匮乏的时代,人工采摘番茄耗时耗力,生产成本很高,应大力发展智能采摘设备,提高采摘效率。针对番茄采摘过程中,番茄果实目标识别不准确以及无法对番茄串采摘点定位的问题,开展

AI Studio

2219人浏览 · 2022-05-27 08:35:23

AI Studio · 2022-05-27 08:35:23 发布

转自AI Studio，原文链接：【AI Workshop】PaddleDetection番茄目标果检测 - 飞桨AI Studio

1、项目内容

随着日常饮食的逐渐丰富,以番茄为原料的食品种类也在不断地增多。在当下劳动力日渐匮乏的时代,人工采摘番茄耗时耗力,生产成本很高,应大力发展智能采摘设备,提高采摘效率。针对番茄采摘过程中,番茄果实目标识别不准确以及无法对番茄串采摘点定位的问题,开展基于深度学习的温室内番茄果实目标识别,可以准确地识别出重叠的番茄果实，希望通过视频监控->目标检测->智能采摘的方式智能、高效的完成此任务。

为解决以上问题，我们选用飞桨目标检测开发套件PaddleDetection来完成，PaddleDetection提供了非常丰富的目标检测模型，项目要求模型精度高、速度达标，本项目采用多种模型，yolov3、ppyolov2和ppyoloe训练进行对比

PaddleDetection为基于飞桨PaddlePaddle的端到端目标检测套件，内置30+模型算法及250+预训练模型，覆盖目标检测、实例分割、跟踪、关键点检测等方向，其中包括服务器端和移动端高精度、轻量级产业级SOTA模型、冠军方案和学术前沿算法，并提供配置化的网络模块组件、十余种数据增强策略和损失函数等高阶优化支持和多种部署方案，在打通数据处理、模型开发、训练、压缩、部署全流程的基础上，提供丰富的案例及教程，加速算法产业落地应用。

提供目标检测、实例分割、多目标跟踪、关键点检测等多种能力应用场景覆盖工业、智慧城市、安防、交通、零售、医疗等十余种行业

模型丰富: 包含目标检测、实例分割、人脸检测、关键点检测、多目标跟踪等250+个预训练模型，涵盖多种全球竞赛冠军方案。
使用简洁：模块化设计，解耦各个网络组件，开发者轻松搭建、试用各种检测模型及优化策略，快速得到高性能、定制化的算法。
端到端打通: 从数据增强、组网、训练、压缩、部署端到端打通，并完备支持云端/边缘端多架构、多设备部署。
高性能: 基于飞桨的高性能内核，模型训练速度及显存占用优势明显。支持FP16训练, 支持多机训练。

最终模型效果展示如下：

视频地址

2、环境要求

PaddlePaddle = 2.2.2
Python = 3.7
PaddleDetection = 2.4
paddlex =2.1.0

3、环境配置

In [ ]

# 下载PaddleDetection代码
# 只有第一次运行本项目时需要执行
!git clone https://gitee.com/paddlepaddle/PaddleDetection

Cloning into 'PaddleDetection'...
remote: Enumerating objects: 23983, done.
remote: Counting objects: 100% (4453/4453), done.
remote: Compressing objects: 100% (2058/2058), done.
Receiving objects:  20% (4876/23983), 7.46 MiB | 714.00 KiB/s

In [ ]

# 下载依赖
# 每次启动项目后都需要先执行
!pip install -r PaddleDetection/requirements.txt

In [ ]

# 更换当前路径 编译安装paddledet
%cd PaddleDetection
!python setup.py install
#  返回主目录
%cd

In [ ]

#  查看gpu版本号
!cat /usr/local/cuda/version.txt

CUDA Version 10.1.243

In [ ]

#  下载对应的版本
!python -m pip install paddlepaddle-gpu==2.3.0.post101 -f https://www.paddlepaddle.org.cn/whl/linux/mkl/avx/stable.html

4、数据集准备

本项目使用的数据集是：番茄采摘数据集该数据集已加载至本环境中，位于：data/data142431/archive (5).zip 本基线系统使用的数据格式是PascalVOC格式
先将data/data142431/目录下archive (5).zip重命名为tomato.zip

In [ ]

# 解压数据
!unzip -o -q -d  /home/aistudio/work /home/aistudio/data/data142431/tomato.zip
#  重命名annotations文件夹名称 
!mv work/annotations work/Annotations
#  重命名images文件夹名称 
!mv work/images work/JPEGImages

In [ ]

#  安装paddlex:为了方便切分数据集
!pip install paddlex==2.1.0

In [ ]

# 切分数据集
!paddlex --split_dataset --format VOC --dataset_dir work --val_value 0.2 --test_value 0.1

划分完成后，该数据集下会生成labels.txt, train_list.txt, val_list.txt和test_list.txt，分别存储类别信息，训练样本列表，验证样本列表，测试样本列表。如下图所示：

将voc数据集转换为coco格式

In [ ]

# 训练集
%cd PaddleDetection/
'''
params
    dataset_type: 原数据格式
    voc_anno_dir: xml标注文件夹
    voc_anno_list: 训练集列表
    voc_label_list: 类别标签
    voc_out_name: 输出json文件
'''
!python tools/x2coco.py \
        --dataset_type voc \
        --voc_anno_dir /home/aistudio/work \
        --voc_anno_list /home/aistudio/work/train_list.txt \
        --voc_label_list /home/aistudio/work/labels.txt \
        --voc_out_name /home/aistudio/work/coco_train.json

[Errno 2] No such file or directory: 'PaddleDetection/'
/home/aistudio/PaddleDetection
Start converting !
100%|██████████████████████████████████████| 627/627 [00:00<00:00, 10028.52it/s]

In [ ]

# 验证集
!python tools/x2coco.py \
       --dataset_type voc \
       --voc_anno_dir /home/aistudio/work/ \
       --voc_anno_list /home/aistudio/work/val_list.txt \
       --voc_label_list /home/aistudio/work/labels.txt \
       --voc_out_name /home/aistudio/work/coco_val.json

Start converting !
100%|██████████████████████████████████████| 179/179 [00:00<00:00, 11010.93it/s]

In [ ]

# 测试集
!python tools/x2coco.py \
       --dataset_type voc \
       --voc_anno_dir /home/aistudio/work/ \
       --voc_anno_list /home/aistudio/work/test_list.txt \
       --voc_label_list /home/aistudio/work/labels.txt \
       --voc_out_name /home/aistudio/work/coco_test.json

Start converting !
100%|█████████████████████████████████████████| 89/89 [00:00<00:00, 8498.61it/s]

In [ ]

%cd

/home/aistudio

5. 模型选择

套件结构概览

Architectures

Backbones

Components

Data Augmentation

Object Detection
- Faster RCNN
- FPN
- Cascade-RCNN
- Libra RCNN
- Hybrid Task RCNN
- PSS-Det
- RetinaNet
- YOLOv3
- YOLOv4
- PP-YOLOv1/v2
- PP-YOLO-Tiny
- SSD
- CornerNet-Squeeze
- FCOS
- TTFNet
- PP-PicoDet
- DETR
- Deformable DETR
- Swin Transformer
- Sparse RCNN
Instance Segmentation
- Mask RCNN
- SOLOv2
Face Detection
- FaceBoxes
- BlazeFace
- BlazeFace-NAS
Multi-Object-Tracking
- JDE
- FairMOT
- DeepSort
KeyPoint-Detection
- HRNet
- HigherHRNet

ResNet(&vd)
ResNeXt(&vd)
SENet
Res2Net
HRNet
Hourglass
CBNet
GCNet
DarkNet
CSPDarkNet
VGG
MobileNetv1/v3
GhostNet
Efficientnet
BlazeNet

Common
- Sync-BN
- Group Norm
- DCNv2
- Non-local

KeyPoint
- DarkPose

FPN
- BiFPN
- BFP
- HRFPN
- ACFPN

Loss
- Smooth-L1
- GIoU/DIoU/CIoU
- IoUAware

Post-processing
- SoftNMS
- MatrixNMS

Speed
- FP16 training
- Multi-machine training

Resize
Lighting
Flipping
Expand
Crop
Color Distort
Random Erasing
Mixup
Mosaic
Cutmix
Grid Mask
Auto Augment
Random Perspective

PaddleDetection 提供了非常丰富的目标检测模型，这里我们选择yolov3、ppyolov2和ppyoloe训练进行对比以ppyoloe配置为例

6. 模型训练

通常一个项目模型完整的落地流程可以总结为如下6个步骤，其中需要根据评估和预测的结果，对模型进行反复的优化和再训练：

也可以参考 PaddleDetecion 提供的快速上手指南。另外，在实际项目的推进过程中，可以根据实际情况决定是否采用模型压缩的相关手段。

6.1 ppyoloe 的配置文件

ppyoloe_crn_s_300e_coco 的配置文件，由 1 个入口配置文件，和5个相关联的子配置文件组成。入口配置文件位于：

PaddleDetection/configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml

6.1.1 数据集配置文件 coco_detection.yml

设置数据集的配置信息。根据本案例的情况，请按照如下内容进行修改

metric: COCO
num_classes: 2

TrainDataset:
  !COCODataSet
    image_dir: JPEGImages
    anno_path: coco_train.json
    dataset_dir: /home/aistudio/work
    data_fields: ['image', 'gt_bbox', 'gt_class', 'is_crowd']

EvalDataset:
  !COCODataSet
    image_dir: JPEGImages
    anno_path: coco_val.json
    dataset_dir: /home/aistudio/work

TestDataset:
  !ImageFolder
    anno_path: coco_test.json # also support txt (like VOC's label_list.txt)
    dataset_dir: /home/aistudio/work # if set, anno_path will be 'dataset_dir/anno_path'

6.1.2 运行时配置文件 runtime.yml

用于设置运行时的参数，主要包括：

use_gpu:  是否使用GPU训练
use_xpu:  是否使用XPU训练
log_iter:  显示训练信息的间隔
save_dir:  模型保存路径
snapshot_epoch: 保存模型的间隔
# Exporting the model: 与导出模型相关的设置

这里我们暂且保留默认值，不做修改即可。

6.1.3 模型网络参数 ppyoloe_crn.yml

用于设置模型的网络参数，也包括预训练集的加载，这里为了可以快速开始实际训练，我们也暂时保留默认的参数，不做修改。

6.1.4 训练优化参数 optimizer_300e.yml

主要说明了学习率和优化器的配置。其中比较重要的参数是训练轮数 epoch 和学习率 base_lr。同样，我们暂时不在这里修改，稍后再设置。

6.1.5 数据读取器配置参数ppyoloe_reader.yml

主要说明了在训练时读取数据集的配置参数，其中比较重要的有：

sample_transforms / batch_transforms: 数据增强算子
batch_size:  批量大小
worker_num:  并发加载子进程数
resize:  读取后的预处理大小

6.1.6 修改入口配置文件PaddleDetection/configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml

这是控制模型训练的主配置文件，其中设置的参数会覆盖掉子配置文件中的相关参数。在主配置文件内集中修改参数，可以更方便的修改训练参数，避免要修改的参数过于分散。

按如下内容修改主配置文件的内容：

_BASE_: [
  '../datasets/coco_detection.yml',
  '../runtime.yml',
  './_base_/optimizer_300e.yml',
  './_base_/ppyoloe_crn.yml',
  './_base_/ppyoloe_reader.yml',
]

log_iter: 100
snapshot_epoch: 2
weights: output/ppyoloe_crn_s_300e_coco/model_final

pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/CSPResNetb_s_pretrained.pdparams
depth_mult: 0.33
width_mult: 0.50

TrainReader:
  batch_size: 12

LearningRate:
  base_lr: 0.001

epoch: 200

LearningRate:
  base_lr: 0.025
  schedulers:
    - !CosineDecay
      max_epochs: 360
    - !LinearWarmup
      start_factor: 0.
      epochs: 5

OptimizerBuilder:
  optimizer:
    momentum: 0.9
    type: Momentum
  regularizer:
    factor: 0.0005
    type: L2

其中：

Batch Size指模型在训练过程中，前向计算一次(即为一个step)所用到的样本数量
如若使用多卡训练， batch_size会均分到各张卡上（因此需要让batch size整除卡数）
Batch Size跟机器的显存/内存高度相关，batch_size越高，所消耗的显存/内存就越高
PP-YOLOE模型训练过程中使用8 GPUs进行混合精度训练，如果GPU卡数或者batch size发生了改变，需要按照公式 lrnew = lrdefault * (batch_sizenew * GPU_numbernew) / (batch_sizedefault * GPU_numberdefault) 调整学习率。

6.2 开始训练

In [ ]

#  yolov3_darknet53训练数据
!export CUDA_VISIBLE_DEVICES=0
!python  PaddleDetection/tools/train.py -c PaddleDetection/configs/yolov3/yolov3_darknet53_270e_voc.yml

In [ ]

#  ppyolo/ppyolov2训练数据
!export CUDA_VISIBLE_DEVICES=0
!python  PaddleDetection/tools/train.py -c PaddleDetection/configs/ppyolo/ppyolov2_r50vd_dcn_voc.yml

In [5]

# ppyoloe_crn_s训练数据
!export CUDA_VISIBLE_DEVICES=0 
!python PaddleDetection/tools/train.py -c PaddleDetection/configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml -r output/ppyoloe_crn_s_300e_coco/239.pdparams

7、模型评估

In [ ]

#评估yolov3_darknet53模型
!export CUDA_VISIBLE_DEVICES=0
!python PaddleDetection/tools/eval.py -c PaddleDetection/configs/yolov3/yolov3_darknet53_270e_voc.yml -o weights=output/yolov3_darknet53_270e_voc/model_final.pdparams

/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/tensor/creation.py:130: 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:
W0519 19:02:17.224961 25187 device_context.cc:447] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.2, Runtime API Version: 10.1
W0519 19:02:17.229497 25187 device_context.cc:465] device: 0, cuDNN Version: 7.6.
[05/19 19:02:20] ppdet.utils.checkpoint INFO: Finish loading model weights: output/yolov3_darknet53_270e_voc/model_final.pdparams
[05/19 19:02:20] ppdet.engine INFO: Eval iter: 0
[05/19 19:02:24] ppdet.metrics.metrics INFO: Accumulating evaluatation results...
[05/19 19:02:24] ppdet.metrics.metrics INFO: mAP(0.50, 11point) = 84.04%
[05/19 19:02:24] ppdet.engine INFO: Total sample number: 89, averge FPS: 20.824143558391864

In [ ]

#评估ppyolov2_r50vd模型
!export CUDA_VISIBLE_DEVICES=0
!python PaddleDetection/tools/eval.py -c PaddleDetection/configs/ppyolo/ppyolov2_r50vd_dcn_voc.yml -o weights=output/ppyolov2_r50vd_dcn_voc/model_final.pdparams

/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/tensor/creation.py:130: 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:
W0519 18:39:34.264786 23682 device_context.cc:447] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.2, Runtime API Version: 10.1
W0519 18:39:34.269814 23682 device_context.cc:465] device: 0, cuDNN Version: 7.6.
[05/19 18:39:38] ppdet.utils.checkpoint INFO: Finish loading model weights: output/ppyolov2_r50vd_dcn_voc/model_final.pdparams
[05/19 18:39:38] ppdet.engine INFO: Eval iter: 0
[05/19 18:39:43] ppdet.metrics.metrics INFO: Accumulating evaluatation results...
[05/19 18:39:43] ppdet.metrics.metrics INFO: mAP(0.50, 11point) = 87.50%
[05/19 18:39:43] ppdet.engine INFO: Total sample number: 89, averge FPS: 18.75877683866569

In [7]

#评估ppyoloe_crn_s_300e模型
!export CUDA_VISIBLE_DEVICES=0
!python PaddleDetection/tools/eval.py -c PaddleDetection/configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml -o weights=output/ppyoloe_crn_s_300e_coco/model_final.pdparams

/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:
W0524 20:57:11.320129 26383 gpu_context.cc:278] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.2, Runtime API Version: 10.1
W0524 20:57:11.324745 26383 gpu_context.cc:306] device: 0, cuDNN Version: 7.6.
loading annotations into memory...
Done (t=0.00s)
creating index...
index created!
[05/24 20:57:15] ppdet.utils.checkpoint INFO: Finish loading model weights: output/ppyoloe_crn_s_300e_coco/model_final.pdparams
[05/24 20:57:15] ppdet.engine INFO: Eval iter: 0
[05/24 20:57:22] ppdet.metrics.metrics INFO: The bbox result is saved to bbox.json.
loading annotations into memory...
Done (t=0.00s)
creating index...
index created!
[05/24 20:57:22] ppdet.metrics.coco_utils INFO: Start evaluate...
Loading and preparing results...
DONE (t=0.08s)
creating index...
index created!
Running per image evaluation...
Evaluate annotation type *bbox*
DONE (t=1.94s).
Accumulating evaluation results...
DONE (t=0.17s).
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.473
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.853
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.466
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.306
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.503
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.627
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.128
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.475
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.589
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.467
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.616
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.715
[05/24 20:57:25] ppdet.engine INFO: Total sample number: 179, averge FPS: 25.355280069172952

	yolov3_darknet53	ppyolov2_r50vd	ppyoloe_crn_s_300e
mAP(.50)	84.04%	87.50%	85.30%
FPS	20.82	18.75	25.35

在同等计算量的前提下，评估三个模型
我们可以看到百度自研算法ppyoloe无论是速度还是精度都是非常强大的。

8、模型导出

将模型模型导出保存在./inference_model

In [ ]

#  yolov3_darknet53模型导出
!python PaddleDetection/tools/export_model.py -c PaddleDetection/configs/yolov3/yolov3_darknet53_270e_voc.yml --output_dir=./inference_model -o weights=output/yolov3_darknet53_270e_voc/model_final.pdparams

/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:
[05/15 16:54:05] ppdet.utils.checkpoint INFO: Finish loading model weights: output/yolov3_darknet53_270e_voc/model_final.pdparams
[05/15 16:54:05] ppdet.engine INFO: Export inference config file to ./inference_model/yolov3_darknet53_270e_voc/infer_cfg.yml
W0515 16:54:09.442514 17021 gpu_context.cc:278] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.2, Runtime API Version: 10.1
W0515 16:54:09.442566 17021 gpu_context.cc:306] device: 0, cuDNN Version: 7.6.
[05/15 16:54:19] ppdet.engine INFO: Export model and saved in ./inference_model/yolov3_darknet53_270e_voc

In [ ]

# ppyolov2_r50vd 模型导出
!python PaddleDetection/tools/export_model.py -c PaddleDetection/configs/ppyolo/ppyolov2_r50vd_dcn_voc.yml --output_dir=./inference_model -o weights=output/ppyolov2_r50vd_dcn_voc/model_final.pdparams

/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/tensor/creation.py:130: 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:
[05/19 18:44:34] ppdet.utils.checkpoint INFO: Finish loading model weights: output/ppyolov2_r50vd_dcn_voc/model_final.pdparams
[05/19 18:44:34] ppdet.engine INFO: Export inference config file to ./inference_model/ppyolov2_r50vd_dcn_voc/infer_cfg.yml
W0519 18:44:39.328130 24180 device_context.cc:447] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.2, Runtime API Version: 10.1
W0519 18:44:39.328193 24180 device_context.cc:465] device: 0, cuDNN Version: 7.6.
[05/19 18:44:44] ppdet.engine INFO: Export model and saved in ./inference_model/ppyolov2_r50vd_dcn_voc

In [8]

# ppyoloe 模型导出
!python PaddleDetection/tools/export_model.py -c PaddleDetection/configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml --output_dir=./inference_model -o weights=output/ppyoloe_crn_s_300e_coco/model_final.pdparams

/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:
[05/24 21:01:52] ppdet.utils.checkpoint INFO: Finish loading model weights: output/ppyoloe_crn_s_300e_coco/model_final.pdparams
loading annotations into memory...
Done (t=0.00s)
creating index...
index created!
[05/24 21:01:52] ppdet.engine INFO: Export inference config file to ./inference_model/ppyoloe_crn_s_300e_coco/infer_cfg.yml
W0524 21:01:55.423156 26759 gpu_context.cc:278] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.2, Runtime API Version: 10.1
W0524 21:01:55.423204 26759 gpu_context.cc:306] device: 0, cuDNN Version: 7.6.
[05/24 21:01:58] ppdet.engine INFO: Export model and saved in ./inference_model/ppyoloe_crn_s_300e_coco

9、预测

9.1 图片预测

In [ ]

#  图片预测
!python PaddleDetection/deploy/python/infer.py --model_dir=inference_model/yolov3_darknet53_270e_voc --image_file=work/JPEGImages/tomato87.png --device=GPU

In [ ]

#  图片预测
!python PaddleDetection/deploy/python/infer.py --model_dir=inference_model/ppyolov2_r50vd_dcn_voc --image_file=work/JPEGImages/tomato87.png --device=GPU

9.2设置参数预测

draw_threshold设置为0.6，输出结果保存在在infer_output

In [ ]

!export CUDA_VISIBLE_DEVICES=0 #windows和Mac下不需要执行该命令
!python PaddleDetection/tools/infer.py -c PaddleDetection/configs/ppyolo/ppyolov2_r50vd_dcn_voc.yml \
                    --infer_img=work/JPEGImages/tomato130.png \
                    --output_dir=infer_output/ \
                    --draw_threshold=0.6 \
                    -o weights=output/ppyolov2_r50vd_dcn_voc/model_final.pdparams \
                    --use_vdl=Ture

/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/tensor/creation.py:130: 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:
W0523 18:45:00.956478  2780 device_context.cc:447] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.2, Runtime API Version: 10.1
W0523 18:45:00.961278  2780 device_context.cc:465] device: 0, cuDNN Version: 7.6.
[05/23 18:45:05] ppdet.utils.checkpoint INFO: Finish loading model weights: output/ppyolov2_r50vd_dcn_voc/model_final.pdparams
100%|█████████████████████████████████████████████| 1/1 [00:00<00:00,  4.19it/s]
[05/23 18:45:05] ppdet.engine INFO: Detection bbox results save in infer_output/tomato130.png

In [ ]

!export CUDA_VISIBLE_DEVICES=0 #windows和Mac下不需要执行该命令
!python PaddleDetection/tools/infer.py -c PaddleDetection/configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml \
                    --infer_img=work/JPEGImages/tomato130.png \
                    --output_dir=infer_output/ \
                    --draw_threshold=0.6 \
                    -o weights=output/ppyoloe_crn_s_300e_coco/model_final.pdparams \
                    --use_vdl=Ture

/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:
W0524 17:23:10.575150 30581 gpu_context.cc:278] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.2, Runtime API Version: 10.1
W0524 17:23:10.579721 30581 gpu_context.cc:306] device: 0, cuDNN Version: 7.6.
[05/24 17:23:13] ppdet.utils.checkpoint INFO: Finish loading model weights: output/ppyoloe_crn_s_300e_coco/109.pdparams
loading annotations into memory...
Done (t=0.00s)
creating index...
index created!
100%|█████████████████████████████████████████████| 1/1 [00:00<00:00,  4.89it/s]
[05/24 17:23:13] ppdet.engine INFO: Detection bbox results save in infer_output/tomato130.png

9.3 视频预测

输出视频保存在output/

In [9]

#  视频预测
!python PaddleDetection/deploy/python/infer.py --model_dir=inference_model/ppyoloe_crn_s_300e_coco --video_file=tomato.mp4 --device=GPU

10 活动总结

AI Workshop是面向深度学习爱好者的开源活动，邀请PPDE或领航团技术大牛作为领航团开源导师，手把手指导成员完成开源项目。在这个过程中呢，从无知到认知，学会了很多很多东西，为将来的就业打下基础。同时开始渐渐喜欢上了这个的专业，感受到学习的过程才是最完美的。非常感谢高睿导师的指导，也非常感谢百度飞桨能够提供这样的机会。

关于作者：华北理工大学轻工学院物联网工程专业大二学生付文昌肖一多陈亮