What is relation between MediaExtractor, MediaCodec and MediaMuxer in android SDK?

I'm trying to make some deep learning experiments on android on video samples. And I've got stuck into remuxing videos. I have a couple of questions to arrange information in my head:) I have read some pages: https://vec.io/posts/android-hardware-decoding-with-mediacodec and https://bigflake.com/mediacodec/#ExtractMpegFramesTest but still I have a mess.

My questions:

1. Can I read video with MediaExtractor and then pass data to MediaMuxer to save video in another file? Without using MediaCodec?
2. If I want to modify frames before saving, can I do that without using Surface? Just by modifying ByteBuffer? I assume that I need to decode data from MediaExtractor, then modify content, then encode it to MediaMuxer.
3. Does sample is the same as frame in context of method MediaExtractor::readSampleData ?
4. Do I need to decode sample?

This is a brief description of what each class does:

• MediaExtrator: Extracts encoded video/audio data
• MediaCodec: Depending on how its configured it can be a decoder or an encoder.
• MediaMuxer: Muxes streams of data into an output file.

This is how you pipeline should generally look like:

MediaExtractor -> MediaCodec(As Decoder) -> Your editing -> MediaCodec(As Encoder) -> MediaMuxer

To answer you questions:

1. MediaExtractor will give you encoded data, if you want to do anything with it you will have to decode it using a MediaCodec.
2. It might be possible to do so without a surface but it will be pretty limited. Surfaces is the way to go. You can find more info here: Editing frames and encoding with MediaCodec
3. Sample can be a video frame or an audio sample
4. Yes you do need to decode samples to edit them