OSError：SavedModel 文件不存在于：model\mymodel.h5/{saved_model.pbtxt|saved_model.pb}

Question

I use tensorflow 2.2 with python 3.7 and windows 10. I import a training model with tensorlfow 1.x. When I import with the terminal everything is fine. But when I freeze my code, the file is no longer found.

Here are my files:

main.py
   \ codes
     import.py
     \ model
        __init__.py
        mymodel.h5

My import.py:

from .model import mymodel

from tensorflow.keras.models import load_model
from tensorflow.keras.preprocessing import image

model = load_model(mymodel)

My init .py code:

from os import path
import sys

if getattr(sys, 'frozen', False):
    MODEL_DIRECTORY = path.join(path.dirname(sys.executable), 'model')
else:
    MODEL_DIRECTORY = path.dirname(__file__)
    
    
mymodel= path.join(MODEL_DIRECTORY, 'mymodel.h5')

The error I got:

SavedModel file does not exist at: exe.win-amd64-3.7\model\mymode

Answer 1

It's clear you're getting part of the python binary mixed from the message. sys.executable is a link to this binary... Probably not what you thought.

if getattr(sys, 'frozen', False):
    MODEL_DIRECTORY = path.join(path.dirname(sys.executable), 'model')
else:
    MODEL_DIRECTORY = path.dirname(__file__)

I'm pretty sure you want:

MODEL_DIRECTORY = path.join(path.dirname(__file__), 'codes', 'model', 'mymodel.h5');

Also your directory structure is weird...

|-|main.py
|-|__init__.py
|-|data
| | |-|model
| | | |-|mymodel.h5
|-|code
| |-|__init__.py
| |-|import.py #wtf is an import.py and why is it down here.

__file__ references the current file so if you have this structure and do this from main.py:

model_dir = Path(path.dirname(__file__), "data", "model", "mymodel.h5") this is using Pathlib.Path and os.path.dirname .