vindioai 0.2.4

📖 Quantum Vision (QV) Theory in Deep Learning

Inspired by quantum physics, Quantum Vision (QV) theory introduces a completely new perspective for object recognition using deep learning.

QV takes captured still images of objects and converts them into information wave functions using a deep learning module called the QV Block.

The QV Block can be integrated into:

• Convolutional Neural Networks (CNNs)
• Vision Transformers (ViTs)
• Convolutional Vision Transformers (CvTs)

These integrations produce QV model variants for object classification.

🔬 Key Highlights

• Converts image data into information wave functions.
• Enhances existing vision architectures by integrating the QV Block.
• Demonstrated consistent performance improvements across multiple datasets compared to standalone models.
• Easy Python API for quick experimentation

Python Usage

🛠️ Installation

# Install via pip (will be available soon on PyPI)
pip install vindioai

🚀 Integrating QVBlock into Your Deep Learning Model (Check the CNN, ViT and CvT integration codes for more details)

import vindioai
from vindioai import QVBlock, Init_Freeze_ShiftSubtract_Layers

# default wave is 128. For more computationally efficient training, wave values can be reduced to 64, 32, 16, 8
wave=128

# For small image size (64, 64 , 3) use momentum_magnitude of either [1] or [1, 2].
#[1] creates 4 parallel branches, while [1, 2] created 8 parallel branches

# For large image sizes (224, 224 , 3) use momentum_magnitude of either [2] or [2, 4]
#[2] creates 4 parallel branches, while [2, 4] created 8 parallel branches

momentum_magnitude=np.array([1, 2])

# Get QV Block
QV_block = QVBlock(momentum_magnitude=momentum_magnitude, input_shape=(64, 64, 3), conv_layers=3, waves=wave)

inputs = QV_block.input
infowave = QV_block.output

# QV_block is now ready to be integrated into your deep learning model

#..... Add your layers such as
# QV INFORMATION WAVES FEED TO A CNN ARCHITECTURE OR VIT OR CvT, CNN example below

convafter1 = Conv2D(128, (3, 3), padding='same', activation=None,
             use_bias=True, name='convafter1')(infowave)

#...

output = Dense(10, activation = 'softmax')(flat)

model = Model(inputs=inputs, outputs=output)

# Must initialize and freeze the weight learning for QVBlock shift and subtract conv layers
model=Init_Freeze_ShiftSubtract_Layers(model, wave, momentum_magnitude=momentum_magnitude)

# Now compile your model as usual
model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])

📜 License

Quantum Vision Theory in Deep Learning for Object Recognition is licensed under the AGPL-3.0 license for research and non-commercial use.
You are free to use, modify, and share the code and models under the terms of the AGPL-3.0 license. However, if you create any derivative works or deploy them (including via a network service), you must release your modifications and the full source code under the same AGPL-3.0 license.

Commercial / Closed-source Use
If you want to integrate Quantum Vision Theory into a closed-source or proprietary product without releasing your source code, you must obtain an Enterprise License from the Vindio AI Software Ltd, https://vindioai.com/

📩 Contact: [info@vindioai.com] for Enterprise pricing and terms.

License Summary

Key Point: If you use Quantum Vision in a product or service and do not want to release your own source code under AGPL-3.0,
you must purchase an Enterprise License.

📄 Citation

If you use this work, please cite our paper:

C. Direkoglu and M. Sah, "Quantum Vision Theory in Deep Learning for Object Recognition," IEEE Access, vol. 13, pp. 132194–132208, 2025.
doi: 10.1109/ACCESS.2025.3592037

bibtex @article{direkoglu2025quantum, author={Direkoglu, Cem and Sah, Melike}, title={Quantum Vision Theory in Deep Learning for Object Recognition}, journal={IEEE Access}, volume={13}, pages={132194-132208}, year={2025}, doi={10.1109/ACCESS.2025.3592037} }

🔬 QVBlock Architecture