An Implementation of Transformer in Transformer in TensorFlow for image classification, attention inside local patches

Last update: Jul 25, 2022

Overview

Transformer-in-Transformer

An Implementation of the Transformer in Transformer paper by Han et al. for image classification, attention inside local patches. Transformer in Transformer uses pixel level attention paired with patch level attention for image classification, in TensorFlow.

PyTorch Implementation

Installation

Run the following to install:

pip install tnt-tensorflow

Developing tnt-tensorflow

To install tnt-tensorflow, along with tools you need to develop and test, run the following in your virtualenv:

git clone https://github.com/Rishit-dagli/Transformer-in-Transformer.git
# or clone your own fork

cd tnt
pip install -e .[dev]

Usage

import tensorflow as tf
from tnt import TNT

tnt = TNT(
    image_size=256,  # size of image
    patch_dim=512,  # dimension of patch token
    pixel_dim=24,  # dimension of pixel token
    patch_size=16,  # patch size
    pixel_size=4,  # pixel size
    depth=5,  # depth
    num_classes=1000,  # output number of classes
    attn_dropout=0.1,  # attention dropout
    ff_dropout=0.1,  # feedforward dropout
)

img = tf.random.uniform(shape=[5, 3, 256, 256])
logits = tnt(img) # (5, 1000)

Want to Contribute 🙋‍♂️ ?

Awesome! If you want to contribute to this project, you're always welcome! See Contributing Guidelines. You can also take a look at open issues for getting more information about current or upcoming tasks.

Want to discuss? 💬

Have any questions, doubts or want to present your opinions, views? You're always welcome. You can start discussions.

Citation

@misc{han2021transformer,
      title={Transformer in Transformer}, 
      author={Kai Han and An Xiao and Enhua Wu and Jianyuan Guo and Chunjing Xu and Yunhe Wang},
      year={2021},
      eprint={2103.00112},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

License

Copyright 2020 Rishit Dagli

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

Comments

Add Unit Tests
The tests should check for the rank and shape of the output tensors, the test should override tf.test.TestCase base class.

[x] #15

[x] #16

[x] #18

[x] #17

Feel free to take inspiration from:

https://github.com/Rishit-dagli/Fast-Transformer/blob/main/fast_transformer/test_fast_transformer.py

For parametrization feel free to follow https://stackoverflow.com/a/34094/11878567, can be used in the exact same way with subTest in TensorFlow

enhancement good first issue
opened by Rishit-dagli 3
Update Workflows to run tests
This issue follows #11

Update GitHub Workflows to:

[ ] Run Tests before uploading to PyPI

[ ] Create a workflow to run tests on commits

Feel free to take inspiration from https://github.com/Rishit-dagli/Fast-Transformer/tree/main/.github/workflows
enhancement good first issue
opened by Rishit-dagli 0

Creates an Attention layer

Verify output shapes just from the attention layer:

import tensorflow as tf
Attention(dim=256)(tf.random.normal([3,256,256]))

# <tf.Tensor: shape=(3, 256, 256), dtype=float32,

Closes #3

opened by Rishit-dagli 0

Put together a TNT class

Verify shapes:

tnt = TNT(
    image_size=256,  # size of image
    patch_dim=512,  # dimension of patch token
    pixel_dim=24,  # dimension of pixel token
    patch_size=16,  # patch size
    pixel_size=4,  # pixel size
    depth=5,  # depth
    num_classes=1000,  # output number of classes
    attn_dropout=0.1,  # attention dropout
    ff_dropout=0.1,  # feedforward dropout
)

img = tf.random.uniform(shape=[1, 3, 256, 256])
print(tnt(img).shape)

# (1, 1000)
```

opened by Rishit-dagli 0

Create an Attention layerr

Verify output shapes just from the attention layer:

import tensorflow as tf
Attention(dim=256)(tf.random.normal([3,256,256]))

# <tf.Tensor: shape=(3, 256, 256), dtype=float32,

opened by Rishit-dagli 0

Create a PreNorm layer

Verify output shapes from this layer:

import tensorflow as tf
PreNorm(dim=1, fn=tf.keras.layers.Dense(5))(tf.random.normal([10, 1]))

# <tf.Tensor: shape=(10, 1), dtype=float32,

opened by Rishit-dagli 0

Releases(v0.2.0)

v0.2.0(Feb 2, 2022)
This is an interesting release for the project, including a pre-trained model on ImageNet, reproducibility of paper results, tests, and end-to-end training.

✅ Bug Fixes / Improvements

Create an end-to-end training example demonstrating how to train a TNT model for image classification through a custom training loop on the TF Flowers dataset (#14)

Pre-trained model to reproduce the paper results have been made available (in this release as well as on TensorFlow Hub)

Create an off-the-shelf inference example, that highlights how you can directly use the pre-trained model made available

Unit Tests for the Attention class (#19)

Unit Tests for the main TNT class (#20)

Full Changelog: https://github.com/Rishit-dagli/Transformer-in-Transformer/compare/v0.1.0...v0.2.0
Source code(tar.gz)
Source code(zip)
tnt_s_patch16_224.tar.gz(84.42 MB)
v0.1.0(Dec 3, 2021)
This is the initial release of TNT TensorFlow and implements Transformers in Transformers as a subclassed TensorFlow model.

Classes

Attention: Implements attention as a TensorFlow Keras Layer making some modifications.

PreNorm: Normalize the activations of the previous layer for each given example in a batch independently and apply some function to it, implemented as a TensorFlow Keras Layer.

FeedForward: Create a FeedForward neural net with two Dense layers and GELU activation, implemented as a TensorFlow Keras Layer.

TNT: Implements the Transformers in Transformers model using all the other classes, and converts to logits. Implemented as a TensorFlow Keras Model.

Source code(tar.gz)
Source code(zip)
tnt_s_patch16_224.tar.gz(84.42 MB)

Owner

Rishit Dagli

High School,TEDx,2xTED-Ed speaker | International Speaker | Microsoft Student Ambassador | Mentor, @TFUGMumbai | Organize @KotlinMumbai

GitHub Repository https://pypi.org/project/tnt-tensorflow

Supplementary materials to "Spin-optomechanical quantum interface enabled by an ultrasmall mechanical and optical mode volume cavity" by H. Raniwala, S. Krastanov, M. Eichenfield, and D. R. Englund, 2022

Supplementary materials to "Spin-optomechanical quantum interface enabled by an ultrasmall mechanical and optical mode volume cavity" by H. Raniwala,

1 Jan 17, 2022

This git repo contains the implementation of my ML project on Heart Disease Prediction

Introduction This git repo contains the implementation of my ML project on Heart Disease Prediction. This is a real-world machine learning model/proje

1 Feb 02, 2022

A PyTorch implementation of "Predict then Propagate: Graph Neural Networks meet Personalized PageRank" (ICLR 2019).

APPNP ⠀ A PyTorch implementation of Predict then Propagate: Graph Neural Networks meet Personalized PageRank (ICLR 2019). Abstract Neural message pass

329 Dec 30, 2022

Statistical-Rethinking-with-Python-and-PyMC3 - Python/PyMC3 port of the examples in " Statistical Rethinking A Bayesian Course with Examples in R and Stan" by Richard McElreath

Statistical Rethinking with Python and PyMC3 This repository has been deprecated in favour of this one, please check that repository for updates, for

786 Dec 29, 2022

An Implementation of Transformer in Transformer in TensorFlow for image classification, attention inside local patches

Related tags

Overview

Transformer-in-Transformer

Installation

Developing tnt-tensorflow

Usage

Want to Contribute 🙋‍♂️ ?

Want to discuss? 💬

Citation

License

Comments

Add Unit Tests

Update Workflows to run tests

Creates an Attention layer

Put together a TNT class

Create an Attention layerr

Create a PreNorm layer

Releases(v0.2.0)

v0.2.0(Feb 2, 2022)

✅ Bug Fixes / Improvements

v0.1.0(Dec 3, 2021)

Classes

Owner

Rishit Dagli

Supplementary materials to "Spin-optomechanical quantum interface enabled by an ultrasmall mechanical and optical mode volume cavity" by H. Raniwala, S. Krastanov, M. Eichenfield, and D. R. Englund, 2022

This git repo contains the implementation of my ML project on Heart Disease Prediction

A PyTorch implementation of "Predict then Propagate: Graph Neural Networks meet Personalized PageRank" (ICLR 2019).

Statistical-Rethinking-with-Python-and-PyMC3 - Python/PyMC3 port of the examples in " Statistical Rethinking A Bayesian Course with Examples in R and Stan" by Richard McElreath

Transfer SemanticKITTI labeles into other dataset/sensor formats.

Reproduce ResNet-v2(Identity Mappings in Deep Residual Networks) with MXNet

Implementation of algorithms for continuous control (DDPG and NAF).

Official repository for: Continuous Control With Ensemble DeepDeterministic Policy Gradients

Embeds a story into a music playlist by sorting the playlist so that the order of the music follows a narrative arc.

Split your patch similarly to `git add -p` but supporting multiple buckets

Official repository of the paper 'Essentials for Class Incremental Learning'

Fast SHAP value computation for interpreting tree-based models

TSDF++: A Multi-Object Formulation for Dynamic Object Tracking and Reconstruction

Streaming over lightweight data transformations

This is the repo for Uncertainty Quantification 360 Toolkit.

Implementation of Feedback Transformer in Pytorch

This script runs neural style transfer against the provided content image.

Brain tumor detection using Convolution-Neural Network (CNN)

This is a work in progress reimplementation of Instant Neural Graphics Primitives

Kaggle Feedback Prize - Evaluating Student Writing 15th solution