GitHub Repository: keras-team/keras-io
Path: blob/master/examples/nlp/md/multimodal_entailment.md
³⁵⁰⁸ views

Multimodal entailment

Author: Sayak Paul
Date created: 2021/08/08
Last modified: 2025/01/03
Description: Training a multimodal model for predicting entailment.

View in Colab • GitHub source

Introduction

In this example, we will build and train a model for predicting multimodal entailment. We will be using the multimodal entailment dataset recently introduced by Google Research.

What is multimodal entailment?

On social media platforms, to audit and moderate content we may want to find answers to the following questions in near real-time:

Does a given piece of information contradict the other?
Does a given piece of information imply the other?

In NLP, this task is called analyzing textual entailment. However, that's only when the information comes from text content. In practice, it's often the case the information available comes not just from text content, but from a multimodal combination of text, images, audio, video, etc. Multimodal entailment is simply the extension of textual entailment to a variety of new input modalities.

Requirements

This example requires TensorFlow 2.5 or higher. In addition, TensorFlow Hub and TensorFlow Text are required for the BERT model (Devlin et al.). These libraries can be installed using the following command:

!pip install -q tensorflow_text

``` [[34;49mnotice[1;39;49m][39;49m A new release of pip is available: [31;49m24.0[39;49m -> [32;49m24.3.1 [[34;49mnotice[1;39;49m][39;49m To update, run: [32;49mpip install --upgrade pip

</div>
---
## Imports


```python
from sklearn.model_selection import train_test_split
import matplotlib.pyplot as plt
import pandas as pd
import numpy as np
import random
import math
from skimage.io import imread
from skimage.transform import resize
from PIL import Image
import os

os.environ["KERAS_BACKEND"] = "jax"  # or tensorflow, or torch

import keras
import keras_hub
from keras.utils import PyDataset

Define a label map

label_map = {"Contradictory": 0, "Implies": 1, "NoEntailment": 2}

Collect the dataset

The original dataset is available here. It comes with URLs of images which are hosted on Twitter's photo storage system called the Photo Blob Storage (PBS for short). We will be working with the downloaded images along with additional data that comes with the original dataset. Thanks to Nilabhra Roy Chowdhury who worked on preparing the image data.

image_base_path = keras.utils.get_file(
    "tweet_images",
    "https://github.com/sayakpaul/Multimodal-Entailment-Baseline/releases/download/v1.0.0/tweet_images.tar.gz",
    untar=True,
)

Read the dataset and apply basic preprocessing

df = pd.read_csv(
    "https://github.com/sayakpaul/Multimodal-Entailment-Baseline/raw/main/csvs/tweets.csv"
).iloc[
    0:1000
]  # Resources conservation since these are examples and not SOTA
df.sample(10)

``` .dataframe tbody tr th { vertical-align: top; }

.dataframe thead th { text-align: right; }

</div>
</style>
<table border="1" class="dataframe">
  <thead>
    <tr style="text-align: right;">
      <th></th>
      <th>id_1</th>
      <th>text_1</th>
      <th>image_1</th>
      <th>id_2</th>
      <th>text_2</th>
      <th>image_2</th>
      <th>label</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <th>815</th>
      <td>1370730009921343490</td>
      <td>Sticky bombs are a threat as they have magnets...</td>
      <td>http://pbs.twimg.com/media/EwXOFrgVIAEkfjR.jpg</td>
      <td>1370731764906295307</td>
      <td>Sticky bombs are a threat as they have magnets...</td>
      <td>http://pbs.twimg.com/media/EwXRK_3XEAA6Q6F.jpg</td>
      <td>NoEntailment</td>
    </tr>
    <tr>
      <th>615</th>
      <td>1364119737446395905</td>
      <td>Daily Horoscope for #Cancer 2.23.21 ♊️❤️✨ #Hor...</td>
      <td>http://pbs.twimg.com/media/Eu5Te44VgAIo1jZ.jpg</td>
      <td>1365218087906078720</td>
      <td>Daily Horoscope for #Cancer 2.26.21 ♊️❤️✨ #Hor...</td>
      <td>http://pbs.twimg.com/media/EvI6nW4WQAA4_E_.jpg</td>
      <td>NoEntailment</td>
    </tr>
    <tr>
      <th>624</th>
      <td>1335542260923068417</td>
      <td>The Reindeer Run is back and this year's run i...</td>
      <td>http://pbs.twimg.com/media/Eoi99DyXEAE0AFV.jpg</td>
      <td>1335872932267122689</td>
      <td>Get your red nose and antlers on for the 2020 ...</td>
      <td>http://pbs.twimg.com/media/Eon5Wk7XUAE-CxN.jpg</td>
      <td>NoEntailment</td>
    </tr>
    <tr>
      <th>970</th>
      <td>1345058844439949312</td>
      <td>Participants needed for online survey!\n\nTopi...</td>
      <td>http://pbs.twimg.com/media/Eqqb4_MXcAA-Pvu.jpg</td>
      <td>1361211461792632835</td>
      <td>Participants needed for top-ranked study on Su...</td>
      <td>http://pbs.twimg.com/media/EuPz0GwXMAMDklt.jpg</td>
      <td>NoEntailment</td>
    </tr>
    <tr>
      <th>456</th>
      <td>1379831489043521545</td>
      <td>comission for @NanoBiteTSF \nenjoyed bros and ...</td>
      <td>http://pbs.twimg.com/media/EyVf0_VXMAMtRaL.jpg</td>
      <td>1380660763749142531</td>
      <td>another comission for @NanoBiteTSF \nhope you ...</td>
      <td>http://pbs.twimg.com/media/EykW0iXXAAA2SBC.jpg</td>
      <td>NoEntailment</td>
    </tr>
    <tr>
      <th>917</th>
      <td>1336180735191891968</td>
      <td>(2/10)\n(Seoul Jung-gu) Market cluster -&amp;gt;\n...</td>
      <td>http://pbs.twimg.com/media/EosRFpGVQAIeuYG.jpg</td>
      <td>1356113330536996866</td>
      <td>(3/11)\n(Seoul Dongdaemun-gu) Goshitel cluster...</td>
      <td>http://pbs.twimg.com/media/EtHhj7QVcAAibvF.jpg</td>
      <td>NoEntailment</td>
    </tr>
    <tr>
      <th>276</th>
      <td>1339270210029834241</td>
      <td>Today the message of freedom goes to Kisoro, R...</td>
      <td>http://pbs.twimg.com/media/EpVK3pfXcAAZ5Du.jpg</td>
      <td>1340881971132698625</td>
      <td>Today the message of freedom is going to the p...</td>
      <td>http://pbs.twimg.com/media/EpvDorkXYAEyz4g.jpg</td>
      <td>Implies</td>
    </tr>
    <tr>
      <th>35</th>
      <td>1360186999836200961</td>
      <td>Bitcoin in Argentina - Google Trends https://t...</td>
      <td>http://pbs.twimg.com/media/EuBa3UxXYAMb99_.jpg</td>
      <td>1382778703055228929</td>
      <td>Argentina wants #Bitcoin https://t.co/9lNxJdxX...</td>
      <td>http://pbs.twimg.com/media/EzCbUFNXMAABwPD.jpg</td>
      <td>Implies</td>
    </tr>
    <tr>
      <th>762</th>
      <td>1370824756400959491</td>
      <td>$HSBA.L: The long term trend is positive and t...</td>
      <td>http://pbs.twimg.com/media/EwYl2hPWYAE2niq.png</td>
      <td>1374347458126475269</td>
      <td>Although the technical rating is only medium, ...</td>
      <td>http://pbs.twimg.com/media/ExKpuwrWgAAktg4.png</td>
      <td>NoEntailment</td>
    </tr>
    <tr>
      <th>130</th>
      <td>1373789433607172097</td>
      <td>I've just watched episode S01 | E05 of Ted Las...</td>
      <td>http://pbs.twimg.com/media/ExCuNbDXAAQaPiL.jpg</td>
      <td>1374913509662806016</td>
      <td>I've just watched episode S01 | E06 of Ted Las...</td>
      <td>http://pbs.twimg.com/media/ExSsjRQWgAUVRPz.jpg</td>
      <td>Contradictory</td>
    </tr>
  </tbody>
</table>
</div>



The columns we are interested in are the following:

* `text_1`
* `image_1`
* `text_2`
* `image_2`
* `label`

The entailment task is formulated as the following:

***Given the pairs of (`text_1`, `image_1`) and (`text_2`, `image_2`) do they entail (or
not entail or contradict) each other?***

We have the images already downloaded. `image_1` is downloaded as `id1` as its filename
and `image2` is downloaded as `id2` as its filename. In the next step, we will add two
more columns to `df` - filepaths of `image_1`s and `image_2`s.


```python
images_one_paths = []
images_two_paths = []

for idx in range(len(df)):
    current_row = df.iloc[idx]
    id_1 = current_row["id_1"]
    id_2 = current_row["id_2"]
    extentsion_one = current_row["image_1"].split(".")[-1]
    extentsion_two = current_row["image_2"].split(".")[-1]

    image_one_path = os.path.join(image_base_path, str(id_1) + f".{extentsion_one}")
    image_two_path = os.path.join(image_base_path, str(id_2) + f".{extentsion_two}")

    images_one_paths.append(image_one_path)
    images_two_paths.append(image_two_path)

df["image_1_path"] = images_one_paths
df["image_2_path"] = images_two_paths

# Create another column containing the integer ids of
# the string labels.
df["label_idx"] = df["label"].apply(lambda x: label_map[x])

Dataset visualization


def visualize(idx):
    current_row = df.iloc[idx]
    image_1 = plt.imread(current_row["image_1_path"])
    image_2 = plt.imread(current_row["image_2_path"])
    text_1 = current_row["text_1"]
    text_2 = current_row["text_2"]
    label = current_row["label"]

    plt.subplot(1, 2, 1)
    plt.imshow(image_1)
    plt.axis("off")
    plt.title("Image One")
    plt.subplot(1, 2, 2)
    plt.imshow(image_1)
    plt.axis("off")
    plt.title("Image Two")
    plt.show()

    print(f"Text one: {text_1}")
    print(f"Text two: {text_2}")
    print(f"Label: {label}")


random_idx = random.choice(range(len(df)))
visualize(random_idx)

random_idx = random.choice(range(len(df)))
visualize(random_idx)

``` Text one: World #water day reminds that we should follow the #guidelines to save water for us. This Day is an #opportunity to learn more about water related issues, be #inspired to tell others and take action to make a difference. Just remember, every #drop counts. ```

``` #WorldWaterDay2021 https://t.co/bQ9Hp53qUj Text two: Water is an extremely precious resource without which life would be impossible. We need to ensure that water is used judiciously, this #WorldWaterDay, let us pledge to reduce water wastage and conserve it. ```

``` #WorldWaterDay2021 https://t.co/0KWnd8Kn8r Label: NoEntailment

</div>
    
![png](/img/examples/nlp/multimodal_entailment/multimodal_entailment_14_2.png)
    


<div class="k-default-codeblock">

Text one: 🎧 𝗘𝗣𝗜𝗦𝗢𝗗𝗘 𝟯𝟬: 𝗗𝗬𝗟𝗔𝗡 𝗙𝗜𝗧𝗭𝗦𝗜𝗠𝗢𝗡𝗦

</div>
    
<div class="k-default-codeblock">

Dylan Fitzsimons is a young passionate greyhound supporter.

</div>
    
<div class="k-default-codeblock">

He and @Drakesport enjoy a great chat about everything greyhounds!

</div>
    
<div class="k-default-codeblock">

Listen: https://t.co/B2XgMp0yaO

</div>
    
<div class="k-default-codeblock">

#GoGreyhoundRacing #ThisRunsDeep #TalkingDogs https://t.co/crBiSqHUvp Text two: 🎧 𝗘𝗣𝗜𝗦𝗢𝗗𝗘 𝟯𝟳: 𝗣𝗜𝗢 𝗕𝗔𝗥𝗥𝗬 🎧

</div>
    
<div class="k-default-codeblock">

Well known within greyhound circles, Pio Barry shares some wonderful greyhound racing stories with @Drakesport in this podcast episode.

</div>
    
<div class="k-default-codeblock">

A great chat.

</div>
    
<div class="k-default-codeblock">

Listen: https://t.co/mJTVlPHzp0

</div>
    
<div class="k-default-codeblock">

#TalkingDogs #GoGreyhoundRacing #ThisRunsDeep https://t.co/QbxtCpLcGm Label: NoEntailment

</div>
---
## Train/test split

The dataset suffers from
[class imbalance problem](https://developers.google.com/machine-learning/glossary#class-imbalanced-dataset).
We can confirm that in the following cell.


```python
df["label"].value_counts()

``` label NoEntailment 819 Contradictory 92 Implies 89 Name: count, dtype: int64

</div>
To account for that we will go for a stratified split.


```python
# 10% for test
train_df, test_df = train_test_split(
    df, test_size=0.1, stratify=df["label"].values, random_state=42
)
# 5% for validation
train_df, val_df = train_test_split(
    train_df, test_size=0.05, stratify=train_df["label"].values, random_state=42
)

print(f"Total training examples: {len(train_df)}")
print(f"Total validation examples: {len(val_df)}")
print(f"Total test examples: {len(test_df)}")

``` Total training examples: 855 Total validation examples: 45 Total test examples: 100

</div>
---
## Data input pipeline

Keras Hub provides
[variety of BERT family of models](https://keras.io/keras_hub/presets/).
Each of those models comes with a
corresponding preprocessing layer. You can learn more about these models and their
preprocessing layers from
[this resource](https://www.kaggle.com/models/keras/bert/keras/bert_base_en_uncased/2).

To keep the runtime of this example relatively short, we will use a base_unacased variant of
the original BERT model.

text preprocessing using KerasHub


```python
text_preprocessor = keras_hub.models.BertTextClassifierPreprocessor.from_preset(
    "bert_base_en_uncased",
    sequence_length=128,
)

Run the preprocessor on a sample input

idx = random.choice(range(len(train_df)))
row = train_df.iloc[idx]
sample_text_1, sample_text_2 = row["text_1"], row["text_2"]
print(f"Text 1: {sample_text_1}")
print(f"Text 2: {sample_text_2}")

test_text = [sample_text_1, sample_text_2]
text_preprocessed = text_preprocessor(test_text)

print("Keys           : ", list(text_preprocessed.keys()))
print("Shape Token Ids : ", text_preprocessed["token_ids"].shape)
print("Token Ids       : ", text_preprocessed["token_ids"][0, :16])
print(" Shape Padding Mask     : ", text_preprocessed["padding_mask"].shape)
print("Padding Mask     : ", text_preprocessed["padding_mask"][0, :16])
print("Shape Segment Ids : ", text_preprocessed["segment_ids"].shape)
print("Segment Ids       : ", text_preprocessed["segment_ids"][0, :16])

``` An NVIDIA GPU may be present on this machine, but a CUDA-enabled jaxlib is not installed. Falling back to cpu.

Text 1: The RPF Lohardaga and Hatia Post of Ranchi Division have recovered 02 bags on 20.02.2021 at Station platform and in T/No.08310 Spl. respectively and handed over to their actual owner correctly. @RPF_INDIA https://t.co/bdEBl2egIc Text 2: The RPF Lohardaga and Hatia Post of Ranchi Division have recovered 02 bags on 20.02.2021 at Station platform and in T/No.08310 (JAT-SBP) Spl. respectively and handed over to their actual owner correctly. @RPF_INDIA https://t.co/Q5l2AtA4uq Keys : ['token_ids', 'padding_mask', 'segment_ids'] Shape Token Ids : (2, 128) Token Ids : [ 101 1996 1054 14376 8840 11783 16098 1998 6045 2401 2695 1997 8086 2072 2407 2031] Shape Padding Mask : (2, 128) Padding Mask : [ True True True True True True True True True True True True True True True True] Shape Segment Ids : (2, 128) Segment Ids : [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]

</div>
We will now create `tf.data.Dataset` objects from the dataframes.

Note that the text inputs will be preprocessed as a part of the data input pipeline. But
the preprocessing modules can also be a part of their corresponding BERT models. This
helps reduce the training/serving skew and lets our models operate with raw text inputs.
Follow [this tutorial](https://www.tensorflow.org/text/tutorials/classify_text_with_bert)
to learn more about how to incorporate the preprocessing modules directly inside the
models.


```python

def dataframe_to_dataset(dataframe):
    columns = ["image_1_path", "image_2_path", "text_1", "text_2", "label_idx"]
    ds = UnifiedPyDataset(
        dataframe,
        batch_size=32,
        workers=4,
    )
    return ds

Preprocessing utilities

bert_input_features = ["padding_mask", "segment_ids", "token_ids"]


def preprocess_text(text_1, text_2):
    output = text_preprocessor([text_1, text_2])
    output = {
        feature: keras.ops.reshape(output[feature], [-1])
        for feature in bert_input_features
    }
    return output

Create the final datasets, method adapted from PyDataset doc string.


class UnifiedPyDataset(PyDataset):
    """A Keras-compatible dataset that processes a DataFrame for TensorFlow, JAX, and PyTorch."""

    def __init__(
        self,
        df,
        batch_size=32,
        workers=4,
        use_multiprocessing=False,
        max_queue_size=10,
        **kwargs,
    ):
        """
        Args:
            df: pandas DataFrame with data
            batch_size: Batch size for dataset
            workers: Number of workers to use for parallel loading (Keras)
            use_multiprocessing: Whether to use multiprocessing
            max_queue_size: Maximum size of the data queue for parallel loading
        """
        super().__init__(**kwargs)
        self.dataframe = df
        columns = ["image_1_path", "image_2_path", "text_1", "text_2"]

        # image files
        self.image_x_1 = self.dataframe["image_1_path"]
        self.image_x_2 = self.dataframe["image_1_path"]
        self.image_y = self.dataframe["label_idx"]

        # text files
        self.text_x_1 = self.dataframe["text_1"]
        self.text_x_2 = self.dataframe["text_2"]
        self.text_y = self.dataframe["label_idx"]

        # general
        self.batch_size = batch_size
        self.workers = workers
        self.use_multiprocessing = use_multiprocessing
        self.max_queue_size = max_queue_size

    def __getitem__(self, index):
        """
        Fetches a batch of data from the dataset at the given index.
        """

        # Return x, y for batch idx.
        low = index * self.batch_size
        # Cap upper bound at array length; the last batch may be smaller
        # if the total number of items is not a multiple of batch size.

        high_image_1 = min(low + self.batch_size, len(self.image_x_1))
        high_image_2 = min(low + self.batch_size, len(self.image_x_2))

        high_text_1 = min(low + self.batch_size, len(self.text_x_1))
        high_text_2 = min(low + self.batch_size, len(self.text_x_1))

        # images files
        batch_image_x_1 = self.image_x_1[low:high_image_1]
        batch_image_y_1 = self.image_y[low:high_image_1]

        batch_image_x_2 = self.image_x_2[low:high_image_2]
        batch_image_y_2 = self.image_y[low:high_image_2]

        # text files
        batch_text_x_1 = self.text_x_1[low:high_text_1]
        batch_text_y_1 = self.text_y[low:high_text_1]

        batch_text_x_2 = self.text_x_2[low:high_text_2]
        batch_text_y_2 = self.text_y[low:high_text_2]

        # image number 1 inputs
        image_1 = [
            resize(imread(file_name), (128, 128)) for file_name in batch_image_x_1
        ]
        image_1 = [
            (  # exeperienced some shapes which were different from others.
                np.array(Image.fromarray((img.astype(np.uint8))).convert("RGB"))
                if img.shape[2] == 4
                else img
            )
            for img in image_1
        ]
        image_1 = np.array(image_1)

        # Both text inputs to the model, return a dict for inputs to BertBackbone
        text = {
            key: np.array(
                [
                    d[key]
                    for d in [
                        preprocess_text(file_path1, file_path2)
                        for file_path1, file_path2 in zip(
                            batch_text_x_1, batch_text_x_2
                        )
                    ]
                ]
            )
            for key in ["padding_mask", "token_ids", "segment_ids"]
        }

        # Image number 2 model inputs
        image_2 = [
            resize(imread(file_name), (128, 128)) for file_name in batch_image_x_2
        ]
        image_2 = [
            (  # exeperienced some shapes which were different from others
                np.array(Image.fromarray((img.astype(np.uint8))).convert("RGB"))
                if img.shape[2] == 4
                else img
            )
            for img in image_2
        ]
        # Stack the list comprehension to an nd.array
        image_2 = np.array(image_2)

        return (
            {
                "image_1": image_1,
                "image_2": image_2,
                "padding_mask": text["padding_mask"],
                "segment_ids": text["segment_ids"],
                "token_ids": text["token_ids"],
            },
            # Target lables
            np.array(batch_image_y_1),
        )

    def __len__(self):
        """
        Returns the number of batches in the dataset.
        """
        return math.ceil(len(self.dataframe) / self.batch_size)

Create train, validation and test datasets


def prepare_dataset(dataframe):
    ds = dataframe_to_dataset(dataframe)
    return ds


train_ds = prepare_dataset(train_df)
validation_ds = prepare_dataset(val_df)
test_ds = prepare_dataset(test_df)

Model building utilities

Our final model will accept two images along with their text counterparts. While the images will be directly fed to the model the text inputs will first be preprocessed and then will make it into the model. Below is a visual illustration of this approach:

The model consists of the following elements:

A standalone encoder for the images. We will use a ResNet50V2 pre-trained on the ImageNet-1k dataset for this.
A standalone encoder for the images. A pre-trained BERT will be used for this.

After extracting the individual embeddings, they will be projected in an identical space. Finally, their projections will be concatenated and be fed to the final classification layer.

This is a multi-class classification problem involving the following classes:

NoEntailment
Implies
Contradictory

project_embeddings(), create_vision_encoder(), and create_text_encoder() utilities are referred from this example.

Projection utilities


def project_embeddings(
    embeddings, num_projection_layers, projection_dims, dropout_rate
):
    projected_embeddings = keras.layers.Dense(units=projection_dims)(embeddings)
    for _ in range(num_projection_layers):
        x = keras.ops.nn.gelu(projected_embeddings)
        x = keras.layers.Dense(projection_dims)(x)
        x = keras.layers.Dropout(dropout_rate)(x)
        x = keras.layers.Add()([projected_embeddings, x])
        projected_embeddings = keras.layers.LayerNormalization()(x)
    return projected_embeddings

Vision encoder utilities


def create_vision_encoder(
    num_projection_layers, projection_dims, dropout_rate, trainable=False
):
    # Load the pre-trained ResNet50V2 model to be used as the base encoder.
    resnet_v2 = keras.applications.ResNet50V2(
        include_top=False, weights="imagenet", pooling="avg"
    )
    # Set the trainability of the base encoder.
    for layer in resnet_v2.layers:
        layer.trainable = trainable

    # Receive the images as inputs.
    image_1 = keras.Input(shape=(128, 128, 3), name="image_1")
    image_2 = keras.Input(shape=(128, 128, 3), name="image_2")

    # Preprocess the input image.
    preprocessed_1 = keras.applications.resnet_v2.preprocess_input(image_1)
    preprocessed_2 = keras.applications.resnet_v2.preprocess_input(image_2)

    # Generate the embeddings for the images using the resnet_v2 model
    # concatenate them.
    embeddings_1 = resnet_v2(preprocessed_1)
    embeddings_2 = resnet_v2(preprocessed_2)
    embeddings = keras.layers.Concatenate()([embeddings_1, embeddings_2])

    # Project the embeddings produced by the model.
    outputs = project_embeddings(
        embeddings, num_projection_layers, projection_dims, dropout_rate
    )
    # Create the vision encoder model.
    return keras.Model([image_1, image_2], outputs, name="vision_encoder")

Text encoder utilities


def create_text_encoder(
    num_projection_layers, projection_dims, dropout_rate, trainable=False
):
    # Load the pre-trained BERT BackBone using KerasHub.
    bert = keras_hub.models.BertBackbone.from_preset(
        "bert_base_en_uncased", num_classes=3
    )

    # Set the trainability of the base encoder.
    bert.trainable = trainable

    # Receive the text as inputs.
    bert_input_features = ["padding_mask", "segment_ids", "token_ids"]
    inputs = {
        feature: keras.Input(shape=(256,), dtype="int32", name=feature)
        for feature in bert_input_features
    }

    # Generate embeddings for the preprocessed text using the BERT model.
    embeddings = bert(inputs)["pooled_output"]

    # Project the embeddings produced by the model.
    outputs = project_embeddings(
        embeddings, num_projection_layers, projection_dims, dropout_rate
    )
    # Create the text encoder model.
    return keras.Model(inputs, outputs, name="text_encoder")

Multimodal model utilities


def create_multimodal_model(
    num_projection_layers=1,
    projection_dims=256,
    dropout_rate=0.1,
    vision_trainable=False,
    text_trainable=False,
):
    # Receive the images as inputs.
    image_1 = keras.Input(shape=(128, 128, 3), name="image_1")
    image_2 = keras.Input(shape=(128, 128, 3), name="image_2")

    # Receive the text as inputs.
    bert_input_features = ["padding_mask", "segment_ids", "token_ids"]
    text_inputs = {
        feature: keras.Input(shape=(256,), dtype="int32", name=feature)
        for feature in bert_input_features
    }
    text_inputs = list(text_inputs.values())
    # Create the encoders.
    vision_encoder = create_vision_encoder(
        num_projection_layers, projection_dims, dropout_rate, vision_trainable
    )
    text_encoder = create_text_encoder(
        num_projection_layers, projection_dims, dropout_rate, text_trainable
    )

    # Fetch the embedding projections.
    vision_projections = vision_encoder([image_1, image_2])
    text_projections = text_encoder(text_inputs)

    # Concatenate the projections and pass through the classification layer.
    concatenated = keras.layers.Concatenate()([vision_projections, text_projections])
    outputs = keras.layers.Dense(3, activation="softmax")(concatenated)
    return keras.Model([image_1, image_2, *text_inputs], outputs)


multimodal_model = create_multimodal_model()
keras.utils.plot_model(multimodal_model, show_shapes=True)

You can inspect the structure of the individual encoders as well by setting the expand_nested argument of plot_model() to True. You are encouraged to play with the different hyperparameters involved in building this model and observe how the final performance is affected.

Compile and train the model

multimodal_model.compile(
    optimizer="adam", loss="sparse_categorical_crossentropy", metrics=["accuracy"]
)

history = multimodal_model.fit(train_ds, validation_data=validation_ds, epochs=1)

``` /home/humbulani/tensorflow-env/env/lib/python3.11/site-packages/keras/src/models/functional.py:248: UserWarning: The structure of `inputs` doesn't match the expected structure. Expected: {'padding_mask': 'padding_mask', 'segment_ids': 'segment_ids', 'token_ids': 'token_ids'} Received: inputs=['Tensor(shape=(32, 256))', 'Tensor(shape=(32, 256))', 'Tensor(shape=(32, 256))'] warnings.warn(msg)

/home/humbulani/tensorflow-env/env/lib/python3.11/site-packages/keras/src/models/functional.py:248: UserWarning: The structure of inputs doesn't match the expected structure. Expected: {'padding_mask': 'padding_mask', 'segment_ids': 'segment_ids', 'token_ids': 'token_ids'} Received: inputs=['Tensor(shape=(32, 256))', 'Tensor(shape=(32, 256))', 'Tensor(shape=(32, 256))'] warnings.warn(msg)

</div>
    
  1/27 [37m━━━━━━━━━━━━━━━━━━━━  45:45 106s/step - accuracy: 0.0625 - loss: 1.6335

<div class="k-default-codeblock">

</div>
  2/27 ━[37m━━━━━━━━━━━━━━━━━━━  42:14 101s/step - accuracy: 0.2422 - loss: 1.9508

<div class="k-default-codeblock">

</div>
  3/27 ━━[37m━━━━━━━━━━━━━━━━━━  38:49 97s/step - accuracy: 0.3524 - loss: 2.0126 

<div class="k-default-codeblock">

</div>
  4/27 ━━[37m━━━━━━━━━━━━━━━━━━  37:09 97s/step - accuracy: 0.4284 - loss: 1.9870

<div class="k-default-codeblock">

</div>
  5/27 ━━━[37m━━━━━━━━━━━━━━━━━  35:08 96s/step - accuracy: 0.4815 - loss: 1.9855

<div class="k-default-codeblock">

</div>
  6/27 ━━━━[37m━━━━━━━━━━━━━━━━  31:56 91s/step - accuracy: 0.5210 - loss: 1.9939

<div class="k-default-codeblock">

</div>
  7/27 ━━━━━[37m━━━━━━━━━━━━━━━  29:30 89s/step - accuracy: 0.5512 - loss: 1.9980

<div class="k-default-codeblock">

</div>
  8/27 ━━━━━[37m━━━━━━━━━━━━━━━  27:12 86s/step - accuracy: 0.5750 - loss: 2.0061

<div class="k-default-codeblock">

</div>
  9/27 ━━━━━━[37m━━━━━━━━━━━━━━  25:15 84s/step - accuracy: 0.5956 - loss: 1.9959

<div class="k-default-codeblock">

</div>
 10/27 ━━━━━━━[37m━━━━━━━━━━━━━  23:33 83s/step - accuracy: 0.6120 - loss: 1.9738

<div class="k-default-codeblock">

</div>
 11/27 ━━━━━━━━[37m━━━━━━━━━━━━  22:09 83s/step - accuracy: 0.6251 - loss: 1.9579

<div class="k-default-codeblock">

</div>
 12/27 ━━━━━━━━[37m━━━━━━━━━━━━  20:59 84s/step - accuracy: 0.6357 - loss: 1.9524

<div class="k-default-codeblock">

</div>
 13/27 ━━━━━━━━━[37m━━━━━━━━━━━  19:44 85s/step - accuracy: 0.6454 - loss: 1.9439

<div class="k-default-codeblock">

</div>
 14/27 ━━━━━━━━━━[37m━━━━━━━━━━  18:22 85s/step - accuracy: 0.6540 - loss: 1.9346

<div class="k-default-codeblock">

/home/humbulani/tensorflow-env/env/lib/python3.11/site-packages/keras/src/models/functional.py:248: UserWarning: The structure of inputs doesn't match the expected structure. Expected: {'padding_mask': 'padding_mask', 'segment_ids': 'segment_ids', 'token_ids': 'token_ids'} Received: inputs=['Tensor(shape=(23, 256))', 'Tensor(shape=(23, 256))', 'Tensor(shape=(23, 256))'] warnings.warn(msg)

</div>
 15/27 ━━━━━━━━━━━[37m━━━━━━━━━  16:52 84s/step - accuracy: 0.6621 - loss: 1.9213

<div class="k-default-codeblock">

</div>
 16/27 ━━━━━━━━━━━[37m━━━━━━━━━  15:29 85s/step - accuracy: 0.6693 - loss: 1.9101

<div class="k-default-codeblock">

</div>
 17/27 ━━━━━━━━━━━━[37m━━━━━━━━  14:08 85s/step - accuracy: 0.6758 - loss: 1.9021

<div class="k-default-codeblock">

</div>
 18/27 ━━━━━━━━━━━━━[37m━━━━━━━  12:45 85s/step - accuracy: 0.6819 - loss: 1.8916

<div class="k-default-codeblock">

</div>
 19/27 ━━━━━━━━━━━━━━[37m━━━━━━  11:24 86s/step - accuracy: 0.6874 - loss: 1.8851

<div class="k-default-codeblock">

</div>
 20/27 ━━━━━━━━━━━━━━[37m━━━━━━  10:00 86s/step - accuracy: 0.6925 - loss: 1.8791

<div class="k-default-codeblock">

</div>
 21/27 ━━━━━━━━━━━━━━━[37m━━━━━  8:36 86s/step - accuracy: 0.6976 - loss: 1.8699 

<div class="k-default-codeblock">

</div>
 22/27 ━━━━━━━━━━━━━━━━[37m━━━━  7:11 86s/step - accuracy: 0.7020 - loss: 1.8623

<div class="k-default-codeblock">

</div>
 23/27 ━━━━━━━━━━━━━━━━━[37m━━━  5:46 87s/step - accuracy: 0.7061 - loss: 1.8573

<div class="k-default-codeblock">

</div>
 24/27 ━━━━━━━━━━━━━━━━━[37m━━━  4:20 87s/step - accuracy: 0.7100 - loss: 1.8534

<div class="k-default-codeblock">

</div>
 25/27 ━━━━━━━━━━━━━━━━━━[37m━━  2:54 87s/step - accuracy: 0.7136 - loss: 1.8494

<div class="k-default-codeblock">

</div>
 26/27 ━━━━━━━━━━━━━━━━━━━[37m━  1:27 87s/step - accuracy: 0.7170 - loss: 1.8449

<div class="k-default-codeblock">

</div>
 27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 88s/step - accuracy: 0.7201 - loss: 1.8414  

<div class="k-default-codeblock">

/home/humbulani/tensorflow-env/env/lib/python3.11/site-packages/PIL/Image.py:1054: UserWarning: Palette images with Transparency expressed in bytes should be converted to RGBA images warnings.warn(

/home/humbulani/tensorflow-env/env/lib/python3.11/site-packages/keras/src/models/functional.py:248: UserWarning: The structure of inputs doesn't match the expected structure. Expected: {'padding_mask': 'padding_mask', 'segment_ids': 'segment_ids', 'token_ids': 'token_ids'} Received: inputs=['Tensor(shape=(13, 256))', 'Tensor(shape=(13, 256))', 'Tensor(shape=(13, 256))'] warnings.warn(msg)

</div>
 27/27 ━━━━━━━━━━━━━━━━━━━━ 2508s 92s/step - accuracy: 0.7231 - loss: 1.8382 - val_accuracy: 0.8222 - val_loss: 1.7304


---
## Evaluate the model


```python
_, acc = multimodal_model.evaluate(test_ds)
print(f"Accuracy on the test set: {round(acc * 100, 2)}%.")

``` /home/humbulani/tensorflow-env/env/lib/python3.11/site-packages/PIL/Image.py:1054: UserWarning: Palette images with Transparency expressed in bytes should be converted to RGBA images warnings.warn(

/home/humbulani/tensorflow-env/env/lib/python3.11/site-packages/PIL/Image.py:1054: UserWarning: Palette images with Transparency expressed in bytes should be converted to RGBA images warnings.warn(

</div>
    
 1/4 ━━━━━[37m━━━━━━━━━━━━━━━  5:32 111s/step - accuracy: 0.7812 - loss: 1.9384

<div class="k-default-codeblock">

</div>
 2/4 ━━━━━━━━━━[37m━━━━━━━━━━  2:10 65s/step - accuracy: 0.7969 - loss: 1.8931 

<div class="k-default-codeblock">

</div>
 3/4 ━━━━━━━━━━━━━━━[37m━━━━━  1:05 65s/step - accuracy: 0.8056 - loss: 1.8200

<div class="k-default-codeblock">

/home/humbulani/tensorflow-env/env/lib/python3.11/site-packages/keras/src/models/functional.py:248: UserWarning: The structure of inputs doesn't match the expected structure. Expected: {'padding_mask': 'padding_mask', 'segment_ids': 'segment_ids', 'token_ids': 'token_ids'} Received: inputs=['Tensor(shape=(4, 256))', 'Tensor(shape=(4, 256))', 'Tensor(shape=(4, 256))'] warnings.warn(msg)

</div>
 4/4 ━━━━━━━━━━━━━━━━━━━━ 0s 49s/step - accuracy: 0.8092 - loss: 1.8075  

<div class="k-default-codeblock">

</div>
 4/4 ━━━━━━━━━━━━━━━━━━━━ 256s 49s/step - accuracy: 0.8113 - loss: 1.8000


<div class="k-default-codeblock">

Accuracy on the test set: 82.0%.

</div>
---
## Additional notes regarding training

**Incorporating regularization**:

The training logs suggest that the model is starting to overfit and may have benefitted
from regularization. Dropout ([Srivastava et al.](https://jmlr.org/papers/v15/srivastava14a.html))
is a simple yet powerful regularization technique that we can use in our model.
But how should we apply it here?

We could always introduce Dropout (`keras.layers.Dropout`) in between different layers of the model.
But here is another recipe. Our model expects inputs from two different data modalities.
What if either of the modalities is not present during inference? To account for this,
we can introduce Dropout to the individual projections just before they get concatenated:

```python
vision_projections = keras.layers.Dropout(rate)(vision_projections)
text_projections = keras.layers.Dropout(rate)(text_projections)
concatenated = keras.layers.Concatenate()([vision_projections, text_projections])

Attending to what matters:

Do all parts of the images correspond equally to their textual counterparts? It's likely not the case. To make our model only focus on the most important bits of the images that relate well to their corresponding textual parts we can use "cross-attention":

# Embeddings.
vision_projections = vision_encoder([image_1, image_2])
text_projections = text_encoder(text_inputs)

# Cross-attention (Luong-style).
query_value_attention_seq = keras.layers.Attention(use_scale=True, dropout=0.2)(
    [vision_projections, text_projections]
)
# Concatenate.
concatenated = keras.layers.Concatenate()([vision_projections, text_projections])
contextual = keras.layers.Concatenate()([concatenated, query_value_attention_seq])

To see this in action, refer to this notebook.

Handling class imbalance:

The dataset suffers from class imbalance. Investigating the confusion matrix of the above model reveals that it performs poorly on the minority classes. If we had used a weighted loss then the training would have been more guided. You can check out this notebook that takes class-imbalance into account during model training.

Using only text inputs:

Also, what if we had only incorporated text inputs for the entailment task? Because of the nature of the text inputs encountered on social media platforms, text inputs alone would have hurt the final performance. Under a similar training setup, by only using text inputs we get to 67.14% top-1 accuracy on the same test set. Refer to this notebook for details.

Finally, here is a table comparing different approaches taken for the entailment task:

Type	Standard Cross-entropy	Loss-weighted Cross-entropy	Focal Loss
Multimodal	77.86%	67.86%	86.43%
Only text	67.14%	11.43%	37.86%

You can check out this repository to learn more about how the experiments were conducted to obtain these numbers.

Final remarks

The architecture we used in this example is too large for the number of data points available for training. It's going to benefit from more data.
We used a smaller variant of the original BERT model. Chances are high that with a larger variant, this performance will be improved. TensorFlow Hub provides a number of different BERT models that you can experiment with.
We kept the pre-trained models frozen. Fine-tuning them on the multimodal entailment task would could resulted in better performance.
We built a simple baseline model for the multimodal entailment task. There are various approaches that have been proposed to tackle the entailment problem. This presentation deck from the Recognizing Multimodal Entailment tutorial provides a comprehensive overview.

You can use the trained model hosted on Hugging Face Hub and try the demo on Hugging Face Spaces