Python 如何用经过训练的地标模型预测自己的形象？

使用tensorflow和get loss 0.0022和val_loss 0.0018对landmark进行了训练，具有很好的锐度。总共7000张图片用于训练。现在我想用我自己的图像进行预测，如何将我自己的图像更改为数组，以便使用模型进行预测。下面是用于预测原始测试图像的代码

X.形状=7000,96,96 Y形=7000,8

Ytrain_pred = model.predict(Xtrain)
Ytest_pred = model.predict(Xtest)

n = 0
nrows = 4
ncols = 4
irand=np.random.choice(Ytest.shape[0],nrows*ncols)
fig, ax = plt.subplots(nrows,ncols,sharex=True,sharey=True,figsize=[ncols*2,nrows*2])
for row in range(nrows):
    for col in range(ncols):
        ax[row,col].imshow(Xtest[irand[n],:,:], cmap='gray')
        ax[row,col].scatter(Ytest[irand[n],0::2]*96,Ytest[irand[n],1::2]*96,marker='X',c='r',s=100)
        ax[row,col].scatter(Ytest_pred[irand[n],0::2]*96,Ytest_pred[irand[n],1::2]*96,marker='+',c='b',s=100)
        ax[row,col].set_xticks(())
        ax[row,col].set_yticks(())
        ax[row,col].set_title('image index = %d' %(irand[n]),fontsize=10)
        n += 1

模型摘要

---

Layer (type)                 Output Shape              Param #   
=================================================================
conv1d_6 (Conv1D)            (None, 96, 32)            15392     
_________________________________________________________________
max_pooling1d_6 (MaxPooling1 (None, 48, 32)            0         
_________________________________________________________________
dropout_11 (Dropout)         (None, 48, 32)            0         
_________________________________________________________________
flatten_6 (Flatten)          (None, 1536)              0         
_________________________________________________________________
dense_11 (Dense)             (None, 256)               393472    
_________________________________________________________________
dropout_12 (Dropout)         (None, 256)               0         
_________________________________________________________________
dense_12 (Dense)             (None, 8)                 2056      
=================================================================
Total params: 410,920
Trainable params: 410,920
Non-trainable params: 0
_________________________________________________________________