ONNX 解析打印 initializer

最新推荐文章于 2024-07-01 22:42:36 发布

6小贱

最新推荐文章于 2024-07-01 22:42:36 发布

阅读量1.5k

点赞数 1

分类专栏： ONNX 文章标签： python 机器学习人工智能

本文链接：https://blog.csdn.net/u013947807/article/details/125387953

版权

ONNX 专栏收录该内容

6 篇文章 0 订阅

订阅专栏

import onnx
import sys
import numpy as np

def onnx_datatype_to_npType(data_type):
    if data_type == 1:
        return np.float32
    else:
        raise TypeError("don't support data type")
    
if __name__ == "__main__":
    
    model_name = "test.onnx"
    
    onnx_model = onnx.load(model_name)
    
    graph = onnx_model.graph
    nodes = graph.node
    initializer = graph.initializer
    inputs= graph.input
    outputs = graph.output

    name_lists = ["Upsample_260","Upsample_267","Upsample_274","Upsample_281"]
    
    for i in range(len(initializer)):
        if initializer[i].name in name_lists:
            print(initializer[i].name,"\t",end="")
            dtype = initializer[i].data_type
            params = np.frombuffer(initializer[i].raw_data, dtype=onnx_datatype_to_npType(dtype))
            print(params,end="\n")