How to save obtained model in terms of
SSDMetaArch object into .pbtxt or .ckpt

To save a model represented as an SSDMetaArch object:

• To save as .ckpt:
  Use the TensorFlow checkpointing mechanism (tf.train.Checkpoint). This saves the model weights (parameters).
  
  

• To save as .pbtxt:
  The .pbtxt is the pipeline config file used to define the model (architecture, training settings). It’s usually saved manually or reused from the original config.
  
  

Key Point:
.ckpt stores weights; .pbtxt stores model architecture/config.
To fully preserve the model, save both after training or modification.