Grad-CAMに自作関数を使った際のエラー

前提・実現したいこと

EfficientNetをファインチューニングしてモデルを作り、Grad-CAMによって可視化しようとしています。grad_cam.pyを実行時、ファインチューニングの際自作のoptimizerを用いていて、モデルの読み込み時にエラーが発生しています。エラー箇所に対してエラー文の意味が理解できない状態です。SGD等既存のものを使っている場合は正常に動作します。以下サイトを参考にしています。よろしくお願いします。
リンク内容
 リンク内容
ファインチューニングの際のソースコードtrain.pyも変更箇所を掲載します。

発生している問題・エラーメッセージ

Using TensorFlow backend.
2020-09-07 09:10:43.135048: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcudart.so.10.1'; dlerror: libcudart.so.10.1: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.0/lib64::/usr/local/lib
2020-09-07 09:10:43.135084: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.
2020-09-07 09:10:43.754234: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcuda.so.1
2020-09-07 09:10:43.775738: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-09-07 09:10:43.776617: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1716] Found device 0 with properties: 
pciBusID: 0000:01:00.0 name: TITAN RTX computeCapability: 7.5
coreClock: 1.77GHz coreCount: 72 deviceMemorySize: 23.62GiB deviceMemoryBandwidth: 625.94GiB/s
2020-09-07 09:10:43.776682: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcudart.so.10.1'; dlerror: libcudart.so.10.1: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.0/lib64::/usr/local/lib
2020-09-07 09:10:43.776723: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcublas.so.10'; dlerror: libcublas.so.10: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.0/lib64::/usr/local/lib
2020-09-07 09:10:43.776762: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcufft.so.10'; dlerror: libcufft.so.10: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.0/lib64::/usr/local/lib
2020-09-07 09:10:43.776800: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcurand.so.10'; dlerror: libcurand.so.10: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.0/lib64::/usr/local/lib
2020-09-07 09:10:43.776838: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcusolver.so.10'; dlerror: libcusolver.so.10: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.0/lib64::/usr/local/lib
2020-09-07 09:10:43.776874: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcusparse.so.10'; dlerror: libcusparse.so.10: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.0/lib64::/usr/local/lib
2020-09-07 09:10:43.778640: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcudnn.so.7
2020-09-07 09:10:43.778648: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1753] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.
Skipping registering GPU devices...
2020-09-07 09:10:43.778873: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN)to use the following CPU instructions in performance-critical operations:  AVX2 FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2020-09-07 09:10:43.801388: I tensorflow/core/platform/profile_utils/cpu_utils.cc:104] CPU Frequency: 3600000000 Hz
2020-09-07 09:10:43.802700: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x5593f39cfe30 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2020-09-07 09:10:43.802711: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): Host, Default Version
2020-09-07 09:10:43.803688: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1257] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-09-07 09:10:43.803695: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1263]      
Traceback (most recent call last):

  File "/home/コンピュータ名/デスクトップ/sakagawa/grad_cam_0831.py", line 19, in <module>
    from adam_lr_mult import Adam_lr_mult

  File "/home/コンピュータ名/デスクトップ/sakagawa/adam_lr_mult.py", line 125, in <module>
    adam_with_lr_multipliers = Adam_lr_mult(multipliers=learning_rate_multipliers)

  File "/home/コンピュータ名/デスクトップ/sakagawa/adam_lr_mult.py", line 38, in __init__
    self.lr = K.variable(lr, name='lr')

AttributeError: can't set attribute

該当のソースコード

python
1#grad_cam.py
2
3import efficientnet.keras
4import numpy as np
5import cv2
6
7# 画像用
8from keras.preprocessing.image import array_to_img, img_to_array, load_img
9# モデル読み込み用
10from keras.models import load_model
11# Grad−CAM計算用
12from tensorflow.keras import models
13import tensorflow as tf
14from adam_lr_mult import Adam_lr_mult
15
16IMAGE_SIZE  = (224, 224)
17
18def grad_cam(input_model, x, layer_name):
19    """
20    Args: 
21        input_model(object): モデルオブジェクト
22        x(ndarray): 画像
23        layer_name(string): 畳み込み層の名前
24    Returns:
25        output_image(ndarray): 元の画像に色付けした画像
26    """
27
28    # 画像の前処理
29    # 読み込む画像が1枚なため、次元を増やしておかないとmode.predictが出来ない
30    X = np.expand_dims(x, axis=0)
31    preprocessed_input = X.astype('float32') / 255.0    
32
33    grad_model = models.Model([input_model.inputs], [input_model.get_layer(layer_name).output, input_model.output])
34
35    with tf.GradientTape() as tape:
36        conv_outputs, predictions = grad_model(preprocessed_input)
37        class_idx = np.argmax(predictions[0])
38        loss = predictions[:, class_idx]
39
40    # 勾配を計算
41    output = conv_outputs[0]
42    grads = tape.gradient(loss, conv_outputs)[0]
43
44    gate_f = tf.cast(output > 0, 'float32')
45    gate_r = tf.cast(grads > 0, 'float32')
46
47    guided_grads = gate_f * gate_r * grads
48
49    # 重みを平均化して、レイヤーの出力に乗じる
50    weights = np.mean(guided_grads, axis=(0, 1))
51    cam = np.dot(output, weights)
52
53    # 画像を元画像と同じ大きさにスケーリング
54    cam = cv2.resize(cam, IMAGE_SIZE, cv2.INTER_LINEAR)
55    # ReLUの代わり
56    cam  = np.maximum(cam, 0)
57    # ヒートマップを計算
58    heatmap = cam / cam.max()
59
60    # モノクロ画像に疑似的に色をつける
61    jet_cam = cv2.applyColorMap(np.uint8(255.0*heatmap), cv2.COLORMAP_JET)
62    # RGBに変換
63    rgb_cam = cv2.cvtColor(jet_cam, cv2.COLOR_BGR2RGB)
64    # もとの画像に合成
65    output_image = (np.float32(rgb_cam) + x / 2)  
66
67    return output_image
68
69model_path = '/home/ho2869/デスクトップ/sakagawa/effB0_auto.h5'
70image_path = '/home/ho2869/デスクトップ/sakagawa/1.jpg'
71
72model = load_model(model_path, custom_objects={'Adam_lr_mult': Adam_lr_mult})
73x = img_to_array(load_img(image_path, target_size=IMAGE_SIZE))
74
75array_to_img(x)
76
77target_layer = 'top_conv'
78cam = grad_cam(model, x, target_layer)
79
80array_to_img(cam)
81cam = cv2.resize(cam, (640, 480))
82cv2.imwrite("grad_cam1.jpg",cam)

python
1#train.py
2
3import efficientnet.keras as efn #記事のままではefficientnetがインポートできなかったため書き方を変えた
4from keras.layers import GlobalAveragePooling2D, Dense, Dropout
5from keras.models import Model
6from data_loader import DataLoader
7from adam_lr_mult import Adam_lr_mult
8
9def prepare_new_model(input_shape, class_count):
10    # 学習済みモデルの取り出し
11    feature_extractor = efn.EfficientNetB0(input_shape=input_shape, weights='imagenet', include_top=False) #インポートに合わせて
12    # 犬猫分類器を引っ付ける
13    x = feature_extractor.output
14    x = GlobalAveragePooling2D()(x)
15    x = Dense(512, activation='relu')(x)
16    x = Dropout(rate=0.25)(x)
17    x = Dense(class_count, activation='sigmoid')(x)
18    # 新たなモデルの定義
19    model = Model(inputs=feature_extractor.input, outputs=x)
20    print(model.summary())
21    return model