moganet

Keras MogaNet

---

Summary

• Keras implementation of Github Westlake-AI/MogaNet. Paper PDF 2211.03295 Efficient Multi-order Gated Aggregation Network.
• Model weights ported from official publication.

---

Models

Model	Params	FLOPs	Input	Top1 Acc	Download
MogaNetXtiny	2.96M	806M	224	76.5	moganet_xtiny_imagenet.h5
MogaNetTiny	5.20M	1.11G	224	79.0	moganet_tiny_224_imagenet.h5
	5.20M	1.45G	256	79.6	moganet_tiny_256_imagenet.h5
MogaNetSmall	25.3M	4.98G	224	83.4	moganet_small_imagenet.h5
MogaNetBase	43.7M	9.96G	224	84.2	moganet_base_imagenet.h5
MogaNetLarge	82.5M	15.96G	224	84.6	moganet_large_imagenet.h5

Usage

from keras_cv_attention_models import moganet

# Will download and load pretrained imagenet weights.
mm = moganet.MogaNetXtiny(pretrained="imagenet")

# Run prediction
import tensorflow as tf
from tensorflow import keras
from skimage.data import chelsea
imm = keras.applications.imagenet_utils.preprocess_input(chelsea(), mode='torch') # Chelsea the cat
pred = mm(tf.expand_dims(tf.image.resize(imm, mm.input_shape[1:3]), 0)).numpy()
print(keras.applications.imagenet_utils.decode_predictions(pred)[0])
# [('n02124075', 'Egyptian_cat', 0.6138564), ('n02123045', 'tabby', 0.16214457), ...]

Change input resolution.

from keras_cv_attention_models import moganet
mm = moganet.MogaNetXtiny(input_shape=(112, 193, 3), pretrained="imagenet")
# >>>> Load pretrained from: ~/.keras/models/caformer_small18_224_imagenet.h5

# Run prediction
from skimage.data import chelsea
preds = mm(mm.preprocess_input(chelsea()))
print(mm.decode_predictions(preds))
# [('n02124075', 'Egyptian_cat', 0.5223805), ('n02123045', 'tabby', 0.27944055), ...]

Use dynamic input resolution by set input_shape=(None, None, 3).

from keras_cv_attention_models import moganet
model = moganet.MogaNetTiny(input_shape=(None, None, 3), num_classes=0)

print(model(np.ones([1, 223, 123, 3])).shape)
# (1, 7, 4, 256)
print(model(np.ones([1, 32, 526, 3])).shape)
# (1, 1, 17, 256)

---