https://github.com/celestialtaha/convnext-tf-keras

tf-keras re-implementation of ConvNeXt (https://arxiv.org/abs/2201.03545) from Facebook Research.
https://github.com/celestialtaha/convnext-tf-keras

2020s-convnets classification convnext deep-learning facebook-research image-net keras modernized-conv tensorflow2

Last synced: 24 days ago
JSON representation

tf-keras re-implementation of ConvNeXt (https://arxiv.org/abs/2201.03545) from Facebook Research.

• Host: GitHub
• URL: https://github.com/celestialtaha/convnext-tf-keras
• Owner: celestialtaha
• License: mit
• Created: 2022-01-21T11:17:50.000Z (over 4 years ago)
• Default Branch: main
• Last Pushed: 2022-01-22T01:22:12.000Z (over 4 years ago)
• Last Synced: 2025-07-21T19:26:59.009Z (11 months ago)
• Topics: 2020s-convnets, classification, convnext, deep-learning, facebook-research, image-net, keras, modernized-conv, tensorflow2
• Language: Jupyter Notebook
• Homepage:
• Size: 46.9 KB
• Stars: 3
• Watchers: 1
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# ConvNeXt-tf-keras

tf-keras re-implementation of ConvNeXt [1] from Facebook Research.

![image](ResNet_vs_ConvNeXt.png)

# Usage

There are two ways one can use this implementation:

  * Notebook (.ipynb)

    - Simply run the cells

  * .py file

## Example of using .py version

```python

from main import create_convnext_model

model = create_convnext_model(input_shape=(224, 224, 3), depths=[3, 3, 9, 3], dims=[96, 192, 384, 768], num_classes=1000)

print(model.summary())

model = create_convnext_model(depths=[3, 3, 9, 3], dims=[96, 192, 384, 768], num_classes=1000, drop_path=2., layer_scale_init_value=1e-6)

print(model.summary())

```

# References

Paper:

[1] [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545)

Code:

[2] https://github.com/facebookresearch/ConvNeXt

[3] https://github.com/rishigami/Swin-Transformer-TF