[Fix] Speed up "--resume"

[chtzs]

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers. By the way, if you're not familiar with how to use pre-commit to fix lint issues or add unit tests, please refer to Contributing to OpenMMLab.

Motivation

In the current version, "--resume" will make engine loading data like regular training, but discard all of this data before reaching the specified iteration. Therefore, the time required for resuming will not be much faster than starting a new training session.

Modification

Added a new method called skip_iter , which skips data without triggering data loading by calling the _next_index in the built-in iterator of DataLoader

Checklist

1. Pre-commit or other linting tools are used to fix the potential lint issues.
2. The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
3. If the modification has potential influence on downstream projects, this PR should be tested with downstream projects, like MMDetection or MMPretrain.
4. The documentation has been modified accordingly, like docstring or example tutorials.

[CLAassistant]

All committers have signed the CLA.

[zhouzaida]

Hi @chtzs , thanks for your contribution. There is a similar issue about the --resume feature. Maybe we can continue the dicussion at #1520

[fpshuang]

插眼

[hyalvin]

hi, when will the PR being merged? Is it possible to release this fix in 0.10.5?

[Zoulinx]

Please merge this PR as soon as possible to ensure the resume feature works properly.