If I have trained the model and want to use pytorch to implement data pre-processing, how should I disassemble the code?

[watertianyi]

描述该功能

If I have already trained the model and want to use pytorch to implement data pre-processing, how should I disassemble the code? I found that your pipeline is as follows:

1. The configuration file needs to go through _init_pipeline(config)
   2.preprocess
   3.track(inputs, 'Inference', total=ceil(1))
   4.data_preprocessor(data, False)
   5.extract_feat(data1['inputs'])
   6.head.predict(feat,data1['data_samples'])

The above, especially 2, 3, and 4, are not easy to understand. The encapsulation is special. Can you elaborate on it? Or tell me whether the above five steps can be fully realized through pytorch.

是否希望自己实现该功能？

• 我希望自己来实现这一功能，并向 MMPreTrain 贡献代码！