paddlelite转换基于SLaNet训练的表格模型报错

[1aqi]

在paddlelite==2.13rc0版本能转换官方表格模型SLANet，但是基于SLANet作为预训练模型训练出来的推理模型无法转换（基于PaddlePadlle==3.0Beta-gpu版本 cuda环境为12.4环境进行训练的模型）

报错信息如下：
C:\Users\seatrend.conda\envs\Paddle-Lite\python.exe D:\Project\python\Paddle-Lite\convert_model.py
[libprotobuf ERROR C:\xiaowen01\x86_windows_py_wheel\py37\Paddle-Lite\third-party\protobuf-host\src\google\protobuf\message_lite.cc:119] Can't parse message of type "paddle.framework.proto.ProgramDesc" because it is missing req
uired fields: blocks[1].ops[41].attrs[13].type, blocks[1].ops[44].attrs[13].type, blocks[1].ops[47].attrs[13].type
Loading topology data from ./model/input/SLANet/inference.pdmodel
[libprotobuf ERROR C:\xiaowen01\x86_windows_py_wheel\py37\Paddle-Lite\third-party\protobuf-host\src\google\protobuf\message_lite.cc:119] Can't parse message of type "paddle.framework.proto.ProgramDesc" because it is missing req
uired fields: blocks[1].ops[41].attrs[13].type, blocks[1].ops[44].attrs[13].type, blocks[1].ops[47].attrs[13].type
Loading params data from ./model/input/SLANet/inference.pdiparams

Model is successfully loaded!
[F 8/ 5 15: 6:54.293 ...e-Lite\lite/core/model/general/op_desc.h:123 paddle::lite::general::OpDesc::GetAttr] Check failed: pair.second->second == OpDataTypeTrait::AT: required type is INTS not match the true type
进程已结束，退出代码为 -1073740791 (0xC0000409)

[ddchenhao66]

您好～请问可以提供下模型和推理代码吗？

[1aqi]

好的，模型是
slanet.zip
推理代码如下：
import os
import cv2
from paddleocr import PPStructure, draw_structure_result, save_structure_res
from PIL import Image
table_engine = PPStructure(layout=False,
show_log=True,
use_angle_cls=True,
use_gpu=False,
det_model_dir="inference/ch_PP-OCRv4_det_infer",
rec_model_dir="inference/ch_PP-OCRv4_rec_infer",
table_model_dir="inference/slanet")

save_folder = './output'
img_path = 'ppocr_img/ppocr_img/imgs/img.png'
img = cv2.imread(img_path)
result = table_engine(img)
save_structure_res(result, save_folder, os.path.basename(img_path).split('.')[0])

for line in result:
line.pop('img')
print(line)

font_path = 'doc/fonts/simfang.ttf' # PaddleOCR下提供字体包
image = Image.open(img_path).convert('RGB')
im_show = draw_structure_result(image, result, font_path=font_path)
im_show = Image.fromarray(im_show)
im_show.save('result.jpg')

[ddchenhao66]

hi～查看了Paddle3.0发版代码，涉及framework.proto改动不多，应该不会影响你这个模型的加载，很大可能是模型文件本身存在问题，在Paddle主框架issue里发现了和你类似的问题PaddlePaddle/Paddle#31741 。此外，也可以尝试回退用老版本paddle训练模型。

[1aqi]

基于官方训练流程下的模型https://aistudio.baidu.com/modelsdetail/277?modelId=277
推理模型导出后，可以使用paddlelite转换成nb模型但是仍然提示
C:\Users\seatrend.conda\envs\Paddle-Lite\python.exe D:\Project\python\Paddle-Lite\convert.py
[libprotobuf ERROR C:\xiaowen01\x86_windows_py_wheel\py37\Paddle-Lite\third-part
y\protobuf-host\src\google\protobuf\message_lite.cc:119] Can't parse message of
type "paddle.framework.proto.ProgramDesc" because it is missing required fields:
blocks[1].ops[42].attrs[13].type, blocks[1].ops[44].attrs[13].type
Loading topology data from ./model/input/table_recognition_module/inference.pdmo
del[libprotobuf ERROR C:\xiaowen01\x86_windows_py_wheel\py37\Paddle-Lite\third-part
y\protobuf-host\src\google\protobuf\message_lite.cc:119] Can't parse message of
type "paddle.framework.proto.ProgramDesc" because it is missing required fields:
blocks[1].ops[42].attrs[13].type, blocks[1].ops[44].attrs[13].type
Loading params data from ./model/input/table_recognition_module/inference.pdiparams

1. Model is successfully loaded!
2. Model is optimized and saved into ./model/output/table_recognition_module_opt.nb successfully
   进程已结束，退出代码为 0
   这边估计问题就是在paddle版本太高的问题。