ppmatting 模型导出为onnx时报pool2d不支持export？

[mmxuan18]

请将下面信息填写完整，便于我们快速解决问题，谢谢！

问题描述 使用下面命令转ppmatting的模型为 onnx时报错 paddle2onnx --model_dir . --model_filename model.pdmodel --params_filename model.pdiparams --save_file model.onnx --enable_dev_version True --opset_version 15 [ERROR][Paddle2ONNX][pool2d: pool2d_1.tmp_0] Adaptive only support static input shape. [Paddle2ONNX] Due to the operator: pool2d, this model cannot be exported to ONNX. [ERROR][Paddle2ONNX][pool2d: pool2d_2.tmp_0] Adaptive only support static input shape. [Paddle2ONNX] Due to the operator: pool2d, this model cannot be exported to ONNX.

模型为https://paddleseg.bj.bcebos.com/matting/models/deploy/pp-matting-hrnet_w18-human_1024.zip

更多信息 :

• 用于部署的推理引擎: onnxruntime
• 为什么需要转换为ONNX格式：能使用现有的服务
• Paddle2ONNX版本: paddle-bfloat 0.1.7 paddle2onnx 0.9.8 paddlepaddle 2.3.1 paddleseg 2.6.0
• 你的联系方式(Email/Wechat/Phone): [email protected] 报错截图

其他信息

[jiangjiajun]

[ERROR][Paddle2ONNX][pool2d: pool2d_2.tmp_0] Adaptive only support static input shape. 根据这里的提示，你这里的pool2d是一个AdaptivePooling , 而ONNX无此对应OP，因此需要固定原模型的输入shape, 在部分情况下，可以将AdaptivePooling转为普通Pool2d进行导出

[LuWei6896]

[ERROR][Paddle2ONNX][pool2d: pool2d_2.tmp_0] Adaptive only support static input shape. 根据这里的提示，你这里的pool2d是一个AdaptivePooling , 而ONNX无此对应OP，因此需要固定原模型的输入shape, 在部分情况下，可以将AdaptivePooling转为普通Pool2d进行导出

同问，怎么“固定原模型的输入shape”？ 我用下面操作进行固化：python paddle_infer_shape.py --model_dir D:/CV/dataset/HumanSeg/inference_models/humansegmodel2 --model_filename model.pdmodel --params_filename model.pdiparams --save_dir D:/CV/dataset/HumanSeg/inference_models/staticinput --input_shape_dict="{'x':[-1,3,192,192]}" 但是并没有输出固化后的模型，也没有报错，只输出了一句话： Start to load paddle model...

[lucasjinreal]

@jiangjiajun I am also confused a lot.....

paddle2onnx --model_dir .\human_pp_humansegv2_lite_192x192_inference_model\ --model_filename model.pdmodel --params_filename model.pdiparams --save_file ppsegv2_lite_192x192.onnx --opset_version 11 --input_shape_dict "{'image':[1, 3, 192, 192]}"
2023-04-08 16:05:56 [WARNING]   [Deprecated] The flag `--input_shape_dict` is deprecated, if you need to modify the input shape of PaddlePaddle model, please refer to this tool https://github.com/jiangjiajun/PaddleUtils/tree/main/paddle
[Paddle2ONNX] Start to parse PaddlePaddle model...
[Paddle2ONNX] Model file path: .\human_pp_humansegv2_lite_192x192_inference_model\model.pdmodel
[Paddle2ONNX] Paramters file path: .\human_pp_humansegv2_lite_192x192_inference_model\model.pdiparams
[Paddle2ONNX] Start to parsing Paddle model...
[ERROR][Paddle2ONNX] [pool2d: pool2d_9.tmp_0] Adaptive only support static input shape.
[Paddle2ONNX] Due to the operator: pool2d, this model cannot be exported to ONNX.
[ERROR][Paddle2ONNX] [pool2d: pool2d_10.tmp_0] Adaptive only support static input shape.
[Paddle2ONNX] Due to the operator: pool2d, this model cannot be exported to ONNX.
[ERROR] Model exporting failed, you can report this problem to https://github.com/PaddlePaddle/Paddle2ONNX.git.

[tianji2018]

原因是在调用export.py时没有固定shape。因为pp-matting-hrnet_w18-human_1024的配置文件没有指定固定的输入shape。再调用export.py时，添加参数--input_shape。例子如下： python tools/export.py \ --config configs/pp_liteseg/pp_liteseg_stdc1_cityscapes_1024x512_scale0.5_160k.yml \ --model_path model.pdparams \ --save_dir output/inference_model \ --input_shape 1 3 1024 1024 导出后，再调用paddle2onnx方法转换模型。

[LuWei6896]

嗯 我已经解决了，感谢--发自新浪邮箱客户端 在 4月11日 16:06，tianji2018 @.***> 写道：

原因是在调用export.py时没有固定shape。因为pp-matting-hrnet_w18-human_1024的配置文件没有指定固定的输入shape。再调用export.py时，添加参数--input_shape。例子如下：

python tools/export.py \ --config configs/pp_liteseg/pp_liteseg_stdc1_cityscapes_1024x512_scale0.5_160k.yml \ --model_path model.pdparams \ --save_dir output/inference_model \ --input_shape 1 3 1024 1024

导出后，再调用paddle2onnx方法转换模型。

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

[Liliyaw]

@jiangjiajun 这个地方我也有个问题, 为什么前段时间还能成功转但是最近就报错了。https://bj.bcebos.com/paddlehub/fastdeploy/Portrait_PP_HumanSegV2_Lite_256x144_with_argmax_infer.tgz 这个模型近期是有调整吗？上周还能成功转onnx但是今天再去测试就报错了。 `2024-01-05 05:09:04 (221 KB/s) - ‘Portrait_PP_HumanSegV2_Lite_256x144_with_argmax_infer.tgz’ saved [3563030/3563030]

./Portrait_PP_HumanSegV2_Lite_256x144_with_argmax_infer/ ./Portrait_PP_HumanSegV2_Lite_256x144_with_argmax_infer/model.pdiparams.info ./Portrait_PP_HumanSegV2_Lite_256x144_with_argmax_infer/model.pdiparams ./Portrait_PP_HumanSegV2_Lite_256x144_with_argmax_infer/deploy.yaml ./Portrait_PP_HumanSegV2_Lite_256x144_with_argmax_infer/model.pdmodel Traceback (most recent call last): File "paddle_infer_shape.py", line 82, in import paddle.fluid as fluid ModuleNotFoundError: No module named 'paddle.fluid' [Paddle2ONNX] Start to parse PaddlePaddle model... [Paddle2ONNX] Model file path: /home/ubuntu/Paddle-examples-for-AVH/object_segmentation/model/model.pdmodel [Paddle2ONNX] Paramters file path: /home/ubuntu/Paddle-examples-for-AVH/object_segmentation/model/model.pdiparams [Paddle2ONNX] Start to parsing Paddle model... [Paddle2ONNX] [bilinear_interp_v2: bilinear_interp_v2_0.tmp_0] Requires the minimal opset version of 11. [ERROR][Paddle2ONNX] [pool2d: pool2d_9.tmp_0] Adaptive only support static input shape. [Paddle2ONNX] Due to the operator: pool2d, this model cannot be exported to ONNX. [Paddle2ONNX] [bilinear_interp_v2: bilinear_interp_v2_1.tmp_0] Requires the minimal opset version of 11. [ERROR][Paddle2ONNX] [pool2d: pool2d_10.tmp_0] Adaptive only support static input shape. [Paddle2ONNX] Due to the operator: pool2d, this model cannot be exported to ONNX. [Paddle2ONNX] [bilinear_interp_v2: bilinear_interp_v2_2.tmp_0] Requires the minimal opset version of 11. [Paddle2ONNX] [bilinear_interp_v2: bilinear_interp_v2_3.tmp_0] Requires the minimal opset version of 11. [Paddle2ONNX] [bilinear_interp_v2: bilinear_interp_v2_4.tmp_0] Requires the minimal opset version of 11. [Paddle2ONNX] [bilinear_interp_v2: bilinear_interp_v2_5.tmp_0] Requires the minimal opset version of 11. [Paddle2ONNX] [bilinear_interp_v2: bilinear_interp_v2_6.tmp_0] Requires the minimal opset version of 11. [Paddle2ONNX] [bilinear_interp_v2: bilinear_interp_v2_7.tmp_0] Requires the minimal opset version of 11. [Paddle2ONNX] [bilinear_interp_v2: bilinear_interp_v2_8.tmp_0] Requires the minimal opset version of 11. [Paddle2ONNX] Due to the operator: bilinear_interp_v2, requires opset_version >= 11. [ERROR] Model exporting failed, you can report this problem to https://github.com/PaddlePaddle/Paddle2ONNX.git.`

[wangkd0]

如果调用export.py时，我不想指定input_size，使用paddle2onnx转换时该怎么处理报错的问题