Unofficial implementation of "Swin Transformer: Hierarchical Vision Transformer using Shifted Windows" (https://arxiv.org/abs/2103.14030)

Last update: Dec 29, 2022

Overview

Swin-Transformer-Tensorflow

A direct translation of the official PyTorch implementation of "Swin Transformer: Hierarchical Vision Transformer using Shifted Windows" to TensorFlow 2.

The official Pytorch implementation can be found here.

Introduction:

Swin Transformer (the name Swin stands for Shifted window) is initially described in arxiv, which capably serves as a general-purpose backbone for computer vision. It is basically a hierarchical Transformer whose representation is computed with shifted windows. The shifted windowing scheme brings greater efficiency by limiting self-attention computation to non-overlapping local windows while also allowing for cross-window connection.

Swin Transformer achieves strong performance on COCO object detection (58.7 box AP and 51.1 mask AP on test-dev) and ADE20K semantic segmentation (53.5 mIoU on val), surpassing previous models by a large margin.

Usage:

1. To Run a Pre-trained Swin Transformer

Swin-T:

python main.py --cfg configs/swin_tiny_patch4_window7_224.yaml --include_top 1 --resume 1 --weights_type imagenet_1k

Swin-S:

python main.py --cfg configs/swin_small_patch4_window7_224.yaml --include_top 1 --resume 1 --weights_type imagenet_1k

Swin-B:

python main.py --cfg configs/swin_base_patch4_window7_224.yaml --include_top 1 --resume 1 --weights_type imagenet_1k

The possible options for cfg and weights_type are:

cfg	weights_type	22K model	1K Model
configs/swin_tiny_patch4_window7_224.yaml	imagenet_1k	-	github
configs/swin_small_patch4_window7_224.yaml	imagenet_1k	-	github
configs/swin_base_patch4_window7_224.yaml	imagenet_1k	-	github
configs/swin_base_patch4_window12_384.yaml	imagenet_1k	-	github
configs/swin_base_patch4_window7_224.yaml	imagenet_22kto1k	-	github
configs/swin_base_patch4_window12_384.yaml	imagenet_22kto1k	-	github
configs/swin_large_patch4_window7_224.yaml	imagenet_22kto1k	-	github
configs/swin_large_patch4_window12_384.yaml	imagenet_22kto1k	-	github
configs/swin_base_patch4_window7_224.yaml	imagenet_22k	github	-
configs/swin_base_patch4_window12_384.yaml	imagenet_22k	github	-
configs/swin_large_patch4_window7_224.yaml	imagenet_22k	github	-
configs/swin_large_patch4_window12_384.yaml	imagenet_22k	github	-

2. Create custom models

To create a custom classification model:

import argparse

import tensorflow as tf

from config import get_config
from models.build import build_model

parser = argparse.ArgumentParser('Custom Swin Transformer')

parser.add_argument(
    '--cfg',
    type=str,
    metavar="FILE",
    help='path to config file',
    default="CUSTOM_YAML_FILE_PATH"
)
parser.add_argument(
    '--resume',
    type=int,
    help='Whether or not to resume training from pretrained weights',
    choices={0, 1},
    default=1,
)
parser.add_argument(
    '--weights_type',
    type=str,
    help='Type of pretrained weight file to load including number of classes',
    choices={"imagenet_1k", "imagenet_22k", "imagenet_22kto1k"},
    default="imagenet_1k",
)

args = parser.parse_args()
custom_config = get_config(args, include_top=False)

swin_transformer = tf.keras.Sequential([
    build_model(config=custom_config, load_pretrained=args.resume, weights_type=args.weights_type),
    tf.keras.layers.Dense(CUSTOM_NUM_CLASSES)
)

Model ouputs are logits, so don't forget to include softmax in training/inference!!

You can easily customize the model configs with custom YAML files. Predefined YAML files provided by Microsoft are located in the configs directory.

3. Convert PyTorch pretrained weights into Tensorflow checkpoints

We provide a python script with which we convert official PyTorch weights into Tensorflow checkpoints.

$ python convert_weights.py --cfg config_file --weights the_path_to_pytorch_weights --weights_type type_of_pretrained_weights --output the_path_to_output_tf_weights

TODO:

Translate model code over to TensorFlow
Load PyTorch pretrained weights into TensorFlow model
Write trainer code
Reproduce results presented in paper
- Object Detection
Reproduce training efficiency of official code in TensorFlow

Citations:

@misc{liu2021swin,
      title={Swin Transformer: Hierarchical Vision Transformer using Shifted Windows}, 
      author={Ze Liu and Yutong Lin and Yue Cao and Han Hu and Yixuan Wei and Zheng Zhang and Stephen Lin and Baining Guo},
      year={2021},
      eprint={2103.14030},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

This is an official implementation of our CVPR 2021 paper "Bottom-Up Human Pose Estimation Via Disentangled Keypoint Regression" (https://arxiv.org/abs/2104.02300)

Bottom-Up Human Pose Estimation Via Disentangled Keypoint Regression Introduction In this paper, we are interested in the bottom-up paradigm of estima

367 Dec 27, 2022

Non-Official Pytorch implementation of "Face Identity Disentanglement via Latent Space Mapping" https://arxiv.org/abs/2005.07728 Using StyleGAN2 instead of StyleGAN

Face Identity Disentanglement via Latent Space Mapping - Implement in pytorch with StyleGAN 2 Description Pytorch implementation of the paper Face Ide

58 Dec 24, 2022

Minimal implementation of PAWS (https://arxiv.org/abs/2104.13963) in TensorFlow.

PAWS-TF 🐾 Implementation of Semi-Supervised Learning of Visual Features by Non-Parametrically Predicting View Assignments with Support Samples (PAWS)

43 Jan 8, 2023

A PyTorch implementation of EventProp [https://arxiv.org/abs/2009.08378], a method to train Spiking Neural Networks

Spiking Neural Network training with EventProp This is an unofficial PyTorch implemenation of EventProp, a method to compute exact gradients for Spiki

35 Jul 29, 2022

Pytorch implementation of Distributed Proximal Policy Optimization: https://arxiv.org/abs/1707.02286

Pytorch-DPPO Pytorch implementation of Distributed Proximal Policy Optimization: https://arxiv.org/abs/1707.02286 Using PPO with clip loss (from https

163 Dec 26, 2022

Tensorflow implementation of Semi-supervised Sequence Learning (https://arxiv.org/abs/1511.01432)

Transfer Learning for Text Classification with Tensorflow Tensorflow implementation of Semi-supervised Sequence Learning(https://arxiv.org/abs/1511.01

82 Oct 22, 2022

PyTorch implementation of Asymmetric Siamese (https://arxiv.org/abs/2204.00613)

Asym-Siam: On the Importance of Asymmetry for Siamese Representation Learning This is a PyTorch implementation of the Asym-Siam paper, CVPR 2022: @inp

89 Dec 18, 2022

This repository contains the code used for Predicting Patient Outcomes with Graph Representation Learning (https://arxiv.org/abs/2101.03940).

Predicting Patient Outcomes with Graph Representation Learning This repository contains the code used for Predicting Patient Outcomes with Graph Repre

76 Dec 22, 2022

https://arxiv.org/abs/2102.11005

LogME LogME: Practical Assessment of Pre-trained Models for Transfer Learning How to use Just feed the features f and labels y to the function, and yo

149 Dec 19, 2022

Comments

Custom Swin Transformer: error: unrecognized arguments

parser = argparse.ArgumentParser('Custom Swin Transformer')

parser.add_argument( '--cfg', type=str, metavar="FILE", help='/content/Swin-Transformer-Tensorflow/configs/swin_tiny_patch4_window7_224.yaml', default="CUSTOM_YAML_FILE_PATH" ) parser.add_argument( '--resume', type=int, help=1, choices={0, 1}, default=1, ) parser.add_argument( '--weights_type', type=str, help='imagenet_22k', choices={"imagenet_1k", "imagenet_22k", "imagenet_22kto1k"}, default="imagenet_1k", )

args = parser.parse_args() custom_config = get_config(args, include_top=False)

i am trying to use it but it throws an error below

usage: Custom Swin Transformer [-h] [--cfg FILE] [--resume {0,1}] [--weights_type {imagenet_22kto1k,imagenet_1k,imagenet_22k}] Custom Swin Transformer: error: unrecognized arguments: -f /root/.local/share/jupyter/runtime/kernel-ee309a98-1f20-4bb7-aa12-c2980aea076c.json An exception has occurred, use %tb to see the full traceback.

SystemExit: 2

opened by AliKayhanAtay 1
train dataset

Thank you for Thank you for providing your code. I've been running the pretrained model, and I'd like to know how to learn about custom data from the code you provided and how to transfer learning to custom data using the pretrained model. Thank you.

opened by hoyeoung 1

Unofficial implementation of "Swin Transformer: Hierarchical Vision Transformer using Shifted Windows" (https://arxiv.org/abs/2103.14030)

Related tags

Overview

Swin-Transformer-Tensorflow

Introduction:

Usage:

1. To Run a Pre-trained Swin Transformer

2. Create custom models

3. Convert PyTorch pretrained weights into Tensorflow checkpoints

TODO:

Citations:

You might also like...

This is an official implementation of our CVPR 2021 paper "Bottom-Up Human Pose Estimation Via Disentangled Keypoint Regression" (https://arxiv.org/abs/2104.02300)

Non-Official Pytorch implementation of "Face Identity Disentanglement via Latent Space Mapping" https://arxiv.org/abs/2005.07728 Using StyleGAN2 instead of StyleGAN

Minimal implementation of PAWS (https://arxiv.org/abs/2104.13963) in TensorFlow.

A PyTorch implementation of EventProp [https://arxiv.org/abs/2009.08378], a method to train Spiking Neural Networks

Pytorch implementation of Distributed Proximal Policy Optimization: https://arxiv.org/abs/1707.02286

Tensorflow implementation of Semi-supervised Sequence Learning (https://arxiv.org/abs/1511.01432)

PyTorch implementation of Asymmetric Siamese (https://arxiv.org/abs/2204.00613)

This repository contains the code used for Predicting Patient Outcomes with Graph Representation Learning (https://arxiv.org/abs/2101.03940).

https://arxiv.org/abs/2102.11005

Comments

Custom Swin Transformer: error: unrecognized arguments

train dataset

Releases(v1.0)

v1.0(Jun 27, 2021)

Owner

This project uses Template Matching technique for object detecting by detection of template image over base image.

A Decentralized Omnidirectional Visual-Inertial-UWB State Estimation System for Aerial Swar.

Mae segmentation - Reproduction of semantic segmentation using masked autoencoder (mae)

Implementation of the paper: "SinGAN: Learning a Generative Model from a Single Natural Image"

Exploration & Research into cross-domain MEV. Initial focus on ETH/POLYGON.

PyTorch implementation of our Adam-NSCL algorithm from our CVPR2021 (oral) paper "Training Networks in Null Space for Continual Learning"

PSML: A Multi-scale Time-series Dataset for Machine Learning in Decarbonized Energy Grids

LaneDetectionAndLaneKeeping - Lane Detection And Lane Keeping

Deep Learning Slide Captcha

The official codes of our CVPR2022 paper: A Differentiable Two-stage Alignment Scheme for Burst Image Reconstruction with Large Shift

Semi-supervised learning for object detection

This is the solution for 2nd rank in Kaggle competition: Feedback Prize - Evaluating Student Writing.

[CVPR 2021] A Peek Into the Reasoning of Neural Networks: Interpreting with Structural Visual Concepts

How Do Adam and Training Strategies Help BNNs Optimization? In ICML 2021.

Implementation of UNet on the Joey ML framework

Open source implementation of AceNAS: Learning to Rank Ace Neural Architectures with Weak Supervision of Weight Sharing

A particular navigation route using satellite feed and can help in toll operations & traffic managemen

Pytorch code for ICRA'21 paper: "Hierarchical Cross-Modal Agent for Robotics Vision-and-Language Navigation"

Lex Rosetta: Transfer of Predictive Models Across Languages, Jurisdictions, and Legal Domains

Video lie detector using xgboost - A video lie detector using OpenFace and xgboost