convert failed node:onnx__Reshape_586, op_type is Conv

xiaohui0225 commented 1 year ago

【FastDeploy版本】： fastdeploy-develop、paddle2onnx=1.0.3 、paddlepaddle-gpu=0.0.0.post111
【系统平台】: Linux x64(Ubuntu 20.04)
【硬件】：说明具体硬件型号，如 Nvidia GPU 3090TI
【编译语言】：Python3.8.3

使用fastdeploy压缩yolov8的模型时出现如下问题，应该是在将onnx转paddle时出错了，问题如下：指令： fastdeploy --auto_compress --config_path=。。。/FastDeploy-develop/tools/auto_compression/configs/detection/yolov8_quant_myself.yaml --method='QAT' --save_dir='。。。/FastDeploy-develop/tools/auto_compression/yolov8_ptq_model_test/' WARNING:root:No module named 'parl' /usr/local/anaconda3/envs/paddle_yolov8/lib/python3.8/site-packages/_distutils_hack/init.py:33: UserWarning: Setuptools is replacing distutils. warnings.warn("Setuptools is replacing distutils.") /usr/local/anaconda3/envs/paddle_yolov8/lib/python3.8/site-packages/paddle/fluid/framework.py:477: UserWarning: PaddlePaddle version 2.3.0 or higher is required, but 0.0.0 installed, Maybe you are using a develop version, please make sure the version is good with your code. /usr/local/anaconda3/envs/paddle_yolov8/lib/python3.8/site-packages/paddleslim-0.0.0.dev0-py3.8.egg/paddleslim/quant/reconstruction_quantization.py:179: SyntaxWarning: "is" with a literal. Did you mean "=="? /usr/local/anaconda3/envs/paddle_yolov8/lib/python3.8/site-packages/paddleslim-0.0.0.dev0-py3.8.egg/paddleslim/quant/reconstruction_quantization.py:179: SyntaxWarning: "is" with a literal. Did you mean "=="? Welcome to use FastDeploy Auto Compression Toolkit! ----------- Running Arguments ----------- Distillation: alpha: 1.0 loss: soft_label Global: arch: YOLOv7 format: onnx input_list: ['x2paddle_images'] model_dir: 。。/Desktop/workspace/yolov8/weights/yolov8s.onnx model_filename: model.pdmodel params_filename: model.pdiparams ptq_image_path: 。。/Desktop/workspace/Dateset/NewDataset-aver2/images/val_small_aug ptq_preprocess: yolo_image_preprocess qat_batch_size: 8 qat_image_path: 。。/Desktop/workspace/Dateset/NewDataset-aver2/images/val_small_aug qat_preprocess: yolo_image_preprocess PTQ: calibration_method: avg skip_tensor_list: None Quantization: activation_quantize_type: moving_average_abs_max onnx_format: True quantize_op_types: ['conv2d', 'depthwise_conv2d'] TrainConfig: epochs: 20 learning_rate: T_max: 8000 learning_rate: 3e-05 type: CosineAnnealingDecay optimizer_builder: optimizer: type: SGD weight_decay: 4e-05 train_iter: 80000

2023-01-11 17:18:48,795-INFO: Now translating model from onnx to paddle. model ir_version: 7, op version: 13 Unknown shape for input tensor[tensor name: 'x2paddle_images'] -> shape: ['batch', 3, 640, 640], Please define shape of input here, Note:you can use visualization tools like Netron to check input shape. Shape of Input(e.g. -1,3,224,224), enter 'N' to skip: -1,3,640,640 shape inferencing ... [WARNING] Incomplete symbolic shape inference shape inferenced. Now, onnx2paddle support convert onnx model opset_verison [7, 8, 9, 10, 11, 12, 13, 14, 15], opset_verison of your onnx model is 13. Total nodes: 259 Nodes converting ... Converting node 386 ... 2023-01-11 17:18:57,088-WARNING: convert failed node:onnx__Reshape_586, op_type is Conv 2023-01-11 17:18:57,088-ERROR: x2paddle threw an exception, you can ask for help at: https://github.com/PaddlePaddle/X2Paddle/issues

请问，出现如上问题是因为fastdeploy还不支持yolov8吗？我所使用的yolov8s.onnx : 链接：https://pan.baidu.com/s/1U77yKCFvq9L2dl22Glvmdg 提取码：uew0

wjj19950828 commented 1 year ago

@xiaohui0225 目前问题已经复现哈，这边会排期看下具体问题

wjj19950828 commented 1 year ago

@xiaohui0225 您好，目前问题已经修复，更新下X2Paddle版本

# 确保老版本已经卸载
pip uninstall x2paddle
pip install x2paddle==1.4.0

xiaohui0225 commented 1 year ago

卸载旧版本，安装1.4.0之后依然出错。

ZJX-CV commented 1 year ago

请问最后解决了吗？

xiaohui0225 commented 1 year ago

请问最后解决了吗？

最后我将训练的pth转成了paddle的model。然后成功了。

ZJX-CV commented 1 year ago

请问没有通过onnx作为中间结果吗？您是直接将pytorch模型作为输入，使用pytorch2paddle吗？

---Original--- From: @.> Date: Mon, Jul 10, 2023 16:13 PM To: @.>; Cc: @.**@.>; Subject: Re: [PaddlePaddle/FastDeploy] convert failed node:onnx__Reshape_586,op_type is Conv (Issue #1119)

请问最后解决了吗？

最后我将训练的pth转成了paddle的model。然后成功了。

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

nkhlS141 commented 9 months ago

@xiaohui0225 what solution worked for you?

xiaohui0225 commented 9 months ago

请问没有通过onnx作为中间结果吗？您是直接将pytorch模型作为输入，使用pytorch2paddle吗？ … ---Original--- From: @.> Date: Mon, Jul 10, 2023 16:13 PM To: @.>; Cc: @.**@.>; Subject: Re: [PaddlePaddle/FastDeploy] convert failed node:onnx__Reshape_586,op_type is Conv (Issue #1119) 请问最后解决了吗？最后我将训练的pth转成了paddle的model。然后成功了。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

现将pytorch得到的pth文件转成paddle的格式，然后再用deploy来量化就好啦

xiaohui0225 commented 9 months ago

@xiaohui0225 what solution worked for you?

现将pytorch得到的pth文件转成paddle的格式，然后再用deploy来量化就好啦

ZJX-CV commented 9 months ago

我一般是将pytorch转成onnx，然后使用x2paddle转成paddle模型。有时候报错，有时候不报错

---Original--- From: @.> Date: Wed, Jan 31, 2024 16:05 PM To: @.>; Cc: @.**@.>; Subject: Re: [PaddlePaddle/FastDeploy] convert failed node:onnx__Reshape_586,op_type is Conv (Issue #1119)

请问没有通过onnx作为中间结果吗？您是直接将pytorch模型作为输入，使用pytorch2paddle吗？ … ---Original--- From: @.> Date: Mon, Jul 10, 2023 16:13 PM To: @.>; Cc: @.@.>; Subject: Re: [PaddlePaddle/FastDeploy] convert failed node:onnx__Reshape_586,op_type is Conv (Issue #1119) 请问最后解决了吗？最后我将训练的pth转成了paddle的model。然后成功了。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

现将pytorch得到的pth文件转成paddle的格式，然后再用deploy来量化就好啦

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

PaddlePaddle / FastDeploy

convert failed node:onnx__Reshape_586, op_type is Conv #1119