You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
99 lines
3.3 KiB
99 lines
3.3 KiB
3 years ago
|
#!/usr/bin/env python
|
||
|
|
|
||
|
|
# 目标检测模型Faster R-CNN训练示例脚本
|
||
|
|
# 执行此脚本前，请确认已正确安装PaddleRS库
|
||
|
|
|
||
|
|
import os
|
||
|
|
|
||
|
|
import paddlers as pdrs
|
||
|
|
from paddlers import transforms as T
|
||
|
|
|
||
|
|
# 下载文件存放目录
|
||
|
|
DOWNLOAD_DIR = './data/sarship/'
|
||
|
|
# 数据集存放目录
|
||
|
|
DATA_DIR = './data/sarship/sar_ship_1/'
|
||
|
|
# 训练集`file_list`文件路径
|
||
|
|
TRAIN_FILE_LIST_PATH = './data/sarship/sar_ship_1/train.txt'
|
||
|
|
# 验证集`file_list`文件路径
|
||
|
|
EVAL_FILE_LIST_PATH = './data/sarship/sar_ship_1/valid.txt'
|
||
|
|
# 数据集类别信息文件路径
|
||
|
|
LABEL_LIST_PATH = './data/sarship/sar_ship_1/labels.txt'
|
||
|
|
# 实验目录，保存输出的模型权重和结果
|
||
|
|
EXP_DIR = './output/faster_rcnn/'
|
||
|
|
|
||
|
|
# 下载和解压SAR影像舰船检测数据集
|
||
|
|
sarship_dataset = 'https://paddleseg.bj.bcebos.com/dataset/sar_ship_1.tar.gz'
|
||
|
|
if not os.path.exists(DATA_DIR):
|
||
|
|
pdrs.utils.download_and_decompress(sarship_dataset, path=DOWNLOAD_DIR)
|
||
|
|
|
||
|
|
# 定义训练和验证时使用的数据变换（数据增强、预处理等）
|
||
|
|
# 使用Compose组合多种变换方式。Compose中包含的变换将按顺序串行执行
|
||
|
|
# API说明：https://github.com/PaddleCV-SIG/PaddleRS/blob/develop/docs/apis/transforms.md
|
||
|
|
train_transforms = T.Compose([
|
||
|
|
# 对输入影像施加随机色彩扰动
|
||
|
|
T.RandomDistort(),
|
||
|
|
# 在影像边界进行随机padding
|
||
|
|
T.RandomExpand(),
|
||
|
|
# 随机裁剪，裁块大小在一定范围内变动
|
||
|
|
T.RandomCrop(),
|
||
|
|
# 随机水平翻转
|
||
|
|
T.RandomHorizontalFlip(),
|
||
|
|
# 对batch进行随机缩放，随机选择插值方式
|
||
|
|
T.BatchRandomResize(
|
||
|
|
target_sizes=[320, 352, 384, 416, 448, 480, 512, 544, 576, 608],
|
||
|
|
interp='RANDOM'),
|
||
|
|
# 影像归一化
|
||
|
|
T.Normalize(
|
||
|
|
mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
|
||
|
|
])
|
||
|
|
|
||
|
|
eval_transforms = T.Compose([
|
||
|
|
# 使用双三次插值将输入影像缩放到固定大小
|
||
|
|
T.Resize(
|
||
|
|
target_size=608, interp='CUBIC'),
|
||
|
|
# 验证阶段与训练阶段的归一化方式必须相同
|
||
|
|
T.Normalize(
|
||
|
|
mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
|
||
|
|
])
|
||
|
|
|
||
|
|
# 分别构建训练和验证所用的数据集
|
||
|
|
train_dataset = pdrs.datasets.VOCDetection(
|
||
|
|
data_dir=DATA_DIR,
|
||
|
|
file_list=TRAIN_FILE_LIST_PATH,
|
||
|
|
label_list=LABEL_LIST_PATH,
|
||
|
|
transforms=train_transforms,
|
||
|
|
shuffle=True)
|
||
|
|
|
||
|
|
eval_dataset = pdrs.datasets.VOCDetection(
|
||
|
|
data_dir=DATA_DIR,
|
||
|
|
file_list=EVAL_FILE_LIST_PATH,
|
||
|
|
label_list=LABEL_LIST_PATH,
|
||
|
|
transforms=eval_transforms,
|
||
|
|
shuffle=False)
|
||
|
|
|
||
|
|
# 构建Faster R-CNN模型
|
||
|
|
# 目前已支持的模型请参考：https://github.com/PaddleCV-SIG/PaddleRS/blob/develop/docs/apis/model_zoo.md
|
||
|
|
# 模型输入参数请参考：https://github.com/PaddleCV-SIG/PaddleRS/blob/develop/paddlers/tasks/object_detector.py
|
||
|
|
model = pdrs.tasks.FasterRCNN(num_classes=len(train_dataset.labels))
|
||
|
|
|
||
|
|
# 执行模型训练
|
||
|
|
model.train(
|
||
|
|
num_epochs=10,
|
||
|
|
train_dataset=train_dataset,
|
||
|
|
train_batch_size=4,
|
||
|
|
eval_dataset=eval_dataset,
|
||
|
|
# 每多少个epoch存储一次检查点
|
||
|
|
save_interval_epochs=5,
|
||
|
|
# 每多少次迭代记录一次日志
|
||
|
|
log_interval_steps=4,
|
||
|
|
save_dir=EXP_DIR,
|
||
|
|
# 指定预训练权重
|
||
|
|
pretrain_weights='COCO',
|
||
|
|
# 初始学习率大小
|
||
|
|
learning_rate=0.005,
|
||
|
|
# 学习率预热（learning rate warm-up）步数与初始值
|
||
|
|
warmup_steps=0,
|
||
|
|
warmup_start_lr=0.0,
|
||
|
|
# 是否启用VisualDL日志功能
|
||
|
|
use_vdl=True)
|