mlx-vision / efficientnet_b5-mlxim

efficientnet_b5

An EfficientNet B5 model architecture, pretrained on ImageNet-1K.

Disclaimer: this is a port of the Torchvision model weights to Apple MLX Framework.

See mlx-convert-scripts repo for the conversion script used.

How to use

pip install mlx-image

Here is how to use this model for image classification:

import mlx.core as mx
from mlxim.model import create_model
from mlxim.io import read_rgb
from mlxim.transform import ImageNetTransform
from mlxim.utils.imagenet import IMAGENET2012_CLASSES

transform = ImageNetTransform(train=False, img_size=456)
x = transform(read_rgb("cat.jpg"))
x = mx.array(x)
x = mx.expand_dims(x, 0)

model = create_model("efficientnet_b5")
model.eval()

logits = model(x)
predicted_idx = mx.argmax(logits, axis=-1).item()
predicted_class = list(IMAGENET2012_CLASSES.values())[predicted_idx]

print(f"Predicted class: {predicted_class}")

You can also use the embeds from layer before head:

import mlx.core as mx
from mlxim.model import create_model
from mlxim.io import read_rgb
from mlxim.transform import ImageNetTransform

transform = ImageNetTransform(train=False, img_size=456)
x = transform(read_rgb("cat.jpg"))
x = mx.array(x)
x = mx.expand_dims(x, 0)

# first option
model = create_model("efficientnet_b5", num_classes=0)
model.eval()

embeds = model(x)

# second option
model = create_model("efficientnet_b5")
model.eval()

embeds = model.get_features(x)