0.2.1

Major Features and Improvements

Add arbitrary shape support for MobileViT* and VisionTransformer*

import numpy as np

import kimm

x = np.random.uniform(size=[1, 150, 150, 3])
model = kimm.models.MobileViTXXS(input_shape=[150, 150, 3])
print(model.predict(x, verbose=0).shape)  # [1, 1000]
model = kimm.models.VisionTransformerTiny16(input_shape=[150, 150, 3])
print(model.predict(x, verbose=0).shape)  # [1, 1000]

Add more model weights

• kimm.models.ConvNeXtXLarge
• kimm.models.VisionTransformerLarge16
• kimm.models.VisionTransformerLarge32

Code cleanup and minor improvements

What's Changed

• Add arbitrary shape support for ViT series by @james77777778 in #46

Full Changelog: 0.2.0...0.2.1