Keras Image Models

• Introduction
• Usage
• Installation
• Quickstart
  • Image classification with ImageNet weights
  • An end-to-end fine-tuning example: cats vs. dogs dataset
  • Grad-CAM
• Model Zoo
• License
• Acknowledgements

Introduction

Keras Image Models (kimm) is a collection of image models, blocks and layers written in Keras 3. The goal is to offer SOTA models with pretrained weights in a user-friendly manner.

KIMM is:

• 🚀 A model zoo where almost all models come with pre-trained weights on ImageNet.
• 🧰 Providing APIs to export models to .tflite and .onnx.
• 🔧 Supporting the reparameterization technique.
• ✨ Integrated with feature extraction capability.

Usage

• kimm.list_models
• kimm.models.*.available_feature_keys
• kimm.models.*(...)
• kimm.models.*(..., feature_extractor=True, feature_keys=[...])
• kimm.utils.get_reparameterized_model
• kimm.export.export_tflite
• kimm.export.export_onnx

import keras
import kimm
import numpy as np


# List available models
print(kimm.list_models("mobileone", weights="imagenet"))
# ['MobileOneS0', 'MobileOneS1', 'MobileOneS2', 'MobileOneS3']

# Initialize model with pretrained ImageNet weights
x = keras.random.uniform([1, 224, 224, 3])
model = kimm.models.MobileOneS0()
y = model.predict(x)
print(y.shape)
# (1, 1000)

# Get reparameterized model by kimm.utils.get_reparameterized_model
reparameterized_model = kimm.utils.get_reparameterized_model(model)
y2 = reparameterized_model.predict(x)
np.testing.assert_allclose(
    keras.ops.convert_to_numpy(y), keras.ops.convert_to_numpy(y2), atol=1e-5
)

# Export model to tflite format
kimm.export.export_tflite(reparameterized_model, 224, "model.tflite")

# Export model to onnx format (note: must be "channels_first" format)
# kimm.export.export_onnx(reparameterized_model, 224, "model.onnx")

# List available feature keys of the model class
print(kimm.models.MobileOneS0.available_feature_keys)
# ['STEM_S2', 'BLOCK0_S4', 'BLOCK1_S8', 'BLOCK2_S16', 'BLOCK3_S32']

# Enable feature extraction by setting `feature_extractor=True`
# `feature_keys` can be optionally specified
model = kimm.models.MobileOneS0(
    feature_extractor=True, feature_keys=["BLOCK2_S16", "BLOCK3_S32"]
)
features = model.predict(x)
for feature_name, feature in features.items():
    print(feature_name, feature.shape)
# BLOCK2_S16 (1, 14, 14, 256)
# BLOCK3_S32 (1, 7, 7, 1024)
# TOP (1, 1000)

Installation

pip install keras kimm -U

Quickstart

Image classification using the model pretrained on ImageNet

Using kimm.models.VisionTransformerTiny16:

1/1 ━━━━━━━━━━━━━━━━━━━━ 1s 1s/step
Predicted: [('n02504458', 'African_elephant', 0.6895825), ('n01871265', 'tusker', 0.17934209), ('n02504013', 'Indian_elephant', 0.12927249)]

An end-to-end example: fine-tuning an image classification model on a cats vs. dogs dataset

Using kimm.models.EfficientNetLiteB0:

Reference: Transfer learning & fine-tuning (keras.io)

Grad-CAM

Using kimm.models.MobileViTS:

Reference: Grad-CAM class activation visualization (keras.io)

Model Zoo

The export scripts can be found in tools/convert_*.py.

License

Please refer to timm as this project is built upon it.

kimm Code

The code here is licensed Apache 2.0.

Acknowledgements

Thanks for these awesome projects that were used in kimm

• https://github.com/keras-team/keras
• https://github.com/huggingface/pytorch-image-models

Citing

BibTeX

@misc{rw2019timm,
  author = {Ross Wightman},
  title = {PyTorch Image Models},
  year = {2019},
  publisher = {GitHub},
  journal = {GitHub repository},
  doi = {10.5281/zenodo.4414861},
  howpublished = {\url{https://github.com/rwightman/pytorch-image-models}}
}

@misc{hy2024kimm,
  author = {Hongyu Chiu},
  title = {Keras Image Models},
  year = {2024},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/james77777778/kimm}}
}