TensorFlow saved_model.pb cannot be loaded correctly

Question

I have downloaded a model from Hub. The archive contains TensorFlow 2 saved model and if decompressed shows a file named saved_model.pb and a variables folder that inside has 2 files variables.data-00000-of-00001 and variables.index.

This model seems cannot be passed to tf.keras.models.load_model() I have tried

my_model=tf.saved_model.load('extraction_path') 

but the resulting loaded_model object seems it isn't a normal model object ready to be used with something like my_model=model.predict(image)and the documentation section about the function isn't clear.

What is the proper procedure that should be used with this model format?

Answer 1

The following works for me on Python 3.9:

import tensorflow as tf
import numpy as np
from PIL import Image

extraction_path = "imagenet_efficientnet_v2_imagenet21k_ft1k_m_classification_2/"
test_image_path = "test.png"
# loading model from path
model = tf.saved_model.load(extraction_path)

# load and reshaping input image
image_np = np.array(Image.open(test_image_path))
input_tensor = tf.convert_to_tensor(image_np, tf.float32)
input_tensor = input_tensor[tf.newaxis, ...]

# run detection on test image
detections = model(input_tensor)

print(detections.shape)
# TensorShape([1, 1000])
# prediction confidence for each of the 1000 classes in the imagenet dataset