NijiFlow
NijiFlow is a Python library wrapping TensorFlow and trained NijiNet models. Its goal is to make it super simple to run 2D-3D image classifier without any knowledge about deep learning or TensorFlow.
Installation
NijiFlow can be simply installed from PyPI with pip. You also need to install TensorFlow to make NijiFlow work.
$ pip install --user nijiflow tensorflow
Supported Python versions are 2.7 and 3.4+.
Usage
import nijiflow
image_paths = [...]
classifier = nijiflow.Classifier()
predictions = classifier.classify(image_paths)
for image_path, prediction in zip(image_paths, predictions):
print('%.3f\t%s' % (prediction, image_path))Prediction values are floating point numbers between 0 and 1. They will be >=0.5 for 2D images, <0.5 otherwise.
NijiNet models
NijiFlow is built on top of NijiNets [2]. NijiNets are MobileNet networks [1] trained for 2D-3D image classification. Details of NijiNets are described in an article in SIG2D Letters #1.
NijiFlow contains the NijiNet model based on MobileNet v1 (1.0/224), but the author also provides several trained models with different parameters. They can be downloaded from following links:
| Network | Size | Accuracy | Precision | Recall | Download |
|---|---|---|---|---|---|
| NijiNet (1.0, 224) | 12MB | 99.1% | 99.7% | 98.5% | nijinet_1.0_224.pb |
| NijiNet (1.0, 128) | 12MB | 98.7% | 99.8% | 97.6% | nijinet_1.0_128.pb |
| NijiNet (0.25, 224) | 0.9MB | 98.5% | 99.7% | 97.4% | nijinet_0.25_224.pb |
License
Aapache 2.0
References
[1] A. G. Howard, M. Zhu, B. Chen, D. Kalenichenko, W. Wang, T. Weyand, M. Andreetto, and H. Adam. Mobilenets: Efficient convolutional neural networks for mobile vision applications. CoRR, abs/1704.04861, 2017.
[2] H. Tachibana. NijiFlow: MobileNets γ«εΊγ₯γγ³γ³γγ―γγͺδΊζ¬‘ε η»εε€ε₯ζ©. SIG2D Letters #1, 2017.
