我在 python 环境中的 Tensorflow 中创建了一个 UNet 模型，并通过 tfjs.converters.save_keras_model() 使用 tfjs 保存了它
模型总结如下。
Model: "model_4"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_4 (InputLayer) [(None, 256, 256, 3) 0
__________________________________________________________________________________________________
sequential_43 (Sequential) (None, 128, 128, 64) 3072 input_4[0][0]
__________________________________________________________________________________________________
sequential_44 (Sequential) (None, 64, 64, 128) 131584 sequential_43[1][0]
__________________________________________________________________________________________________
sequential_45 (Sequential) (None, 32, 32, 256) 525312 sequential_44[0][0]
__________________________________________________________________________________________________
sequential_46 (Sequential) (None, 16, 16, 512) 2099200 sequential_45[0][0]
__________________________________________________________________________________________________
sequential_47 (Sequential) (None, 8, 8, 512) 4196352 sequential_46[0][0]
__________________________________________________________________________________________________
sequential_48 (Sequential) (None, 4, 4, 512) 4196352 sequential_47[0][0]
__________________________________________________________________________________________________
sequential_49 (Sequential) (None, 2, 2, 512) 4196352 sequential_48[0][0]
__________________________________________________________________________________________________
sequential_50 (Sequential) (None, 1, 1, 512) 4196352 sequential_49[0][0]
__________________________________________________________________________________________________
sequential_51 (Sequential) (None, 2, 2, 512) 4196352 sequential_50[0][0]
__________________________________________________________________________________________________
concatenate_16 (Concatenate) (None, 2, 2, 1024) 0 sequential_51[0][0]
sequential_49[0][0]
__________________________________________________________________________________________________
sequential_52 (Sequential) (None, 4, 4, 512) 8390656 concatenate_16[0][0]
__________________________________________________________________________________________________
concatenate_17 (Concatenate) (None, 4, 4, 1024) 0 sequential_52[0][0]
sequential_48[0][0]
__________________________________________________________________________________________________
sequential_53 (Sequential) (None, 8, 8, 512) 8390656 concatenate_17[0][0]
__________________________________________________________________________________________________
concatenate_18 (Concatenate) (None, 8, 8, 1024) 0 sequential_53[0][0]
sequential_47[0][0]
__________________________________________________________________________________________________
sequential_54 (Sequential) (None, 16, 16, 512) 8390656 concatenate_18[0][0]
__________________________________________________________________________________________________
concatenate_19 (Concatenate) (None, 16, 16, 1024) 0 sequential_54[0][0]
sequential_46[0][0]
__________________________________________________________________________________________________
sequential_55 (Sequential) (None, 32, 32, 256) 4195328 concatenate_19[0][0]
__________________________________________________________________________________________________
concatenate_20 (Concatenate) (None, 32, 32, 512) 0 sequential_55[0][0]
sequential_45[0][0]
__________________________________________________________________________________________________
sequential_56 (Sequential) (None, 64, 64, 128) 1049088 concatenate_20[0][0]
__________________________________________________________________________________________________
concatenate_21 (Concatenate) (None, 64, 64, 256) 0 sequential_56[0][0]
sequential_44[0][0]
__________________________________________________________________________________________________
sequential_57 (Sequential) (None, 128, 128, 64) 262400 concatenate_21[0][0]
__________________________________________________________________________________________________
concatenate_22 (Concatenate) (None, 128, 128, 128 0 sequential_57[0][0]
sequential_43[1][0]
__________________________________________________________________________________________________
conv2d_transpose_26 (Conv2DTran (None, 256, 256, 3) 6147 concatenate_22[0][0]
==================================================================================================
Total params: 54,425,859
Trainable params: 54,414,979
Non-trainable params: 10,880
当我检查 model.input_shape 它输出 - (None, 256, 256, 3)
因此，模型定义了输入形状，但是在将其加载到 tfjs 时，错误表明 Sequential 的第一层应该具有 input_shape 或 batch_input_shape
当我从 tfjs 转换后检查 model.json 时，它具有以下第一个输入层的配置
"config":
{
"batch_input_shape": [null, 256, 256, 3],
...
那么，我应该怎么做才能在浏览器上加载 TensorFlowJS 中的模型呢？