https://github.com/920232796/setr-pytorch

Implementation of SETR model, Original paper: Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers.
https://github.com/920232796/setr-pytorch

pytorch setr transformer transformer-segmentation

Last synced: 6 days ago
JSON representation

Implementation of SETR model, Original paper: Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers.

• Host: GitHub
• URL: https://github.com/920232796/setr-pytorch
• Owner: 920232796
• Created: 2021-01-18T07:57:40.000Z (almost 4 years ago)
• Default Branch: master
• Last Pushed: 2021-02-08T10:54:33.000Z (almost 4 years ago)
• Last Synced: 2024-10-31T17:44:32.765Z (13 days ago)
• Topics: pytorch, setr, transformer, transformer-segmentation
• Language: Python
• Homepage:
• Size: 46 MB
• Stars: 129
• Watchers: 6
• Forks: 20
• Open Issues: 10
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        



## SETR - Pytorch

Since the original paper (Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers.) has no official code,I implemented SETR-Progressive UPsampling(SETR-PUP) using pytorch.

Original paper: Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers.

## Vit

The Vit model is also implemented, and you can use it for image classification.

## Usage SETR

```python

from SETR.transformer_seg import SETRModel

import torch 

if __name__ == "__main__":

    net = SETRModel(patch_size=(32, 32), 

                    in_channels=3, 

                    out_channels=1, 

                    hidden_size=1024, 

                    num_hidden_layers=8, 

                    num_attention_heads=16, 

                    decode_features=[512, 256, 128, 64])

    t1 = torch.rand(1, 3, 256, 256)

    print("input: " + str(t1.shape))

    

    # print(net)

    print("output: " + str(net(t1).shape))

```

If the output size is (1, 1, 256, 256), the code runs successfully.

## Usage Vit

```python 

from SETR.transformer_seg import Vit

import torch 

if __name__ == "__main__":

    model = Vit(patch_size=(7, 7), 

                    in_channels=1, 

                    out_class=10, 

                    hidden_size=1024, 

                    num_hidden_layers=1, 

                    num_attention_heads=16)

    print(model)

    t1 = torch.rand(1, 1, 28, 28)

    print("input: " + str(t1.shape))

    print("output: " + str(model(t1).shape))

```

The output shape is (1, 10).

## current examples

1. task_mnist: The simplest example, using the Vit model to classify the minst dataset.

2. task_car_seg: The example is sample segmentation task. data download: https://www.kaggle.com/c/carvana-image-masking-challenge/data

## more

More examples will be updated later.