训练语义分割模型，自动下载deeplabv3预训练模型DeepLabv3p_Xception65_COCO后，但报错不是有效参数。

源自github用户panicwq:
Exception: File output/meter_seg\pretrain\DeepLabv3p_Xception65_COCO is not a paddle parameter file

版本、环境信息：
1）PaddlePaddle版本：paddlepaddle-gpu 2.2.1
2）Paddlex: 1.3.10
3）系统环境：win10 python3.8
4) GPU:GTX1650Ti CUDA10.2 cudnn 8.2.2.26
训练信息
1）单机，单卡

源自github用户cxxly:
您好，辛苦提供下复现该问题最小代码。

源自github用户panicwq:
代码就是paddlex官方案例：工业指针表计读数中训练语义分割模型的代码 train_segmentation.py
`import os

os.environ['CUDA_VISIBLE_DEVICES'] = '0'

import paddlex as pdx
from paddlex.seg import transforms

meter_seg_dataset = 'https://bj.bcebos.com/paddlex/examples/meter_reader/datasets/meter_seg.tar.gz'
pdx.utils.download_and_decompress(meter_seg_dataset, path='./')

train_transforms = transforms.Compose([
transforms.Resize([512, 512]),
transforms.RandomHorizontalFlip(prob=0.5),
transforms.Normalize(),
])

eval_transforms = transforms.Compose([
transforms.Resize([512, 512]),
transforms.Normalize(),
])

train_dataset = pdx.datasets.SegDataset(
data_dir='meter_seg/',
file_list='meter_seg/train.txt',
label_list='meter_seg/labels.txt',
transforms=train_transforms,
shuffle=True)
eval_dataset = pdx.datasets.SegDataset(
data_dir='meter_seg/',
file_list='meter_seg/val.txt',
label_list='meter_seg/labels.txt',
transforms=eval_transforms)

model = pdx.seg.DeepLabv3p(
num_classes=len(train_dataset.labels), backbone='Xception65')
model.train(
num_epochs=20,
train_dataset=train_dataset,
train_batch_size=4,
eval_dataset=eval_dataset,
learning_rate=0.1,
pretrain_weights='COCO',
save_interval_epochs=5,
save_dir='output/meter_seg',
use_vdl=True)`