Merge branch 'release/1.0.2'

README.md

@@ -6,6 +6,12 @@ Based on [Official Pytorch implementation](https://github.com/DingXiaoH/RepLKNet
 
 Supports variable-shape inference.
 
+## Installation
+
+```bash
+pip install tfreplknet
+```
+
 ## Examples
 
 Default usage (without preprocessing):
@@ -42,11 +48,17 @@ with [ImageNet-v2 test set](https://www.tensorflow.org/datasets/catalog/imagenet
 ```python
 import tensorflow as tf
 import tensorflow_datasets as tfds
-from tfreplknet import RepLKNet31B224K1, preprocess_input
+from tfreplknet import RepLKNet31B224K1, RepLKNet31B384K1, preprocess_input
 
 def _prepare(example):
-    image = tf.image.resize(example['image'], (256, 256), method=tf.image.ResizeMethod.BICUBIC, antialias=False)
+    # For RepLKNet31B224K1
+    image = tf.image.resize(example['image'], (256, 256), method=tf.image.ResizeMethod.BICUBIC)
     image = tf.image.central_crop(image, 0.875)
+
+    # For RepLKNet31B384K1
+    # image = tf.image.resize(example['image'], (438, 438), method=tf.image.ResizeMethod.BICUBIC)
+    # image = tf.image.central_crop(image, 0.877)
+
     image = preprocess_input(image)
 
     return image, example['label']
@@ -64,8 +76,8 @@ print(history)
 
 | name | original acc@1 | ported acc@1 | original acc@5 | ported acc@5 |
 | :---: | :---: | :---: | :---: | :---: |
-| RepLKNet31B 224 1K | ? | ? | ? | ? |
-| RepLKNet31B 384 1K | ? | ? | ? | ? |
+| RepLKNet31B 224 1K | 75.29 | 75.13 | 92.60 | 92.88 |
+| RepLKNet31B 384 1K | ? | 76.46 | ? | 93.37 |
 
 ## Citation
 

setup.py

@@ -6,7 +6,7 @@
 
 setup(
     name='tfreplknet',
-    version='1.0.1',
+    version='1.0.2',
     description='Keras (TensorFlow v2) reimplementation of RepLKNet model.',
     long_description=long_description,
     long_description_content_type="text/markdown",

tfreplknet/model.py

@@ -91,20 +91,20 @@ def RepLKNet(filters, kernel_sizes=(31, 29, 27, 13), small_kernel=5, depths=(2,
     x = SamePad(3)(x)
     x = layers.Conv2D(filters[0], 3, strides=2, use_bias=False, name='stem/0/conv')(x)
     x = layers.BatchNormalization(momentum=0.1, epsilon=1.001e-5, name='stem/0/bn')(x)
-    x = layers.ReLU()(x)
+    x = layers.ReLU(name='stem/0/relu')(x)
 
     x = layers.DepthwiseConv2D(3, padding='same', use_bias=False, name='stem/1/conv')(x)
     x = layers.BatchNormalization(momentum=0.1, epsilon=1.001e-5, name='stem/1/bn')(x)
-    x = layers.ReLU()(x)
+    x = layers.ReLU(name='stem/1/relu')(x)
 
     x = layers.Conv2D(filters[0], 1, use_bias=False, name='stem/2/conv')(x)
     x = layers.BatchNormalization(momentum=0.1, epsilon=1.001e-5, name='stem/2/bn')(x)
-    x = layers.ReLU()(x)
+    x = layers.ReLU(name='stem/2/relu')(x)
 
     x = SamePad(3)(x)
     x = layers.DepthwiseConv2D(3, strides=2, use_bias=False, name='stem/3/conv')(x)
     x = layers.BatchNormalization(momentum=0.1, epsilon=1.001e-5, name='stem/3/bn')(x)
-    x = layers.ReLU()(x)
+    x = layers.ReLU(name='stem/3/relu')(x)
 
     path_drops = np.linspace(0., path_drop, sum(depths))
 
@@ -118,12 +118,12 @@ def RepLKNet(filters, kernel_sizes=(31, 29, 27, 13), small_kernel=5, depths=(2,
         if not_last:
             x = layers.Conv2D(filters[i + 1], 1, use_bias=False, name=f'transitions/{i}/0/conv')(x)
             x = layers.BatchNormalization(momentum=0.1, epsilon=1.001e-5, name=f'transitions/{i}/0/bn')(x)
-            x = layers.ReLU()(x)
+            x = layers.ReLU(name=f'transitions/{i}/0/relu')(x)
 
             x = SamePad(3)(x)
             x = layers.DepthwiseConv2D(3, strides=2, use_bias=False, name=f'transitions/{i}/1/conv')(x)
             x = layers.BatchNormalization(momentum=0.1, epsilon=1.001e-5, name=f'transitions/{i}/1/bn')(x)
-            x = layers.ReLU()(x)
+            x = layers.ReLU(name=f'transitions/{i}/1/relu')(x)
 
     x = layers.BatchNormalization(momentum=0.1, epsilon=1.001e-5, name='norm')(x)