[Feature] add ORL

[zhaozh10]

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.

Motivation

Add ORL algorithm, including Python files and README.md

Modification

Please briefly describe what modification is made in this PR.
ORL is composed of three stages, which can be concluded as stage 1: Image-level pre-training and knn image ids retrieve, stage 2: Roi_generate, stage 2: Roi_pair_retrieve and stage 3: object-level pre-training

[1] For dataset-related issues,
a. Regist 'SSDataset', 'CorrespondDataset', 'ORLDataset' in mmselfsup/dataset/init.py
b. ORL involves COCO train2017 dataset in stage 1 and stage 3. Add configs/selfsup/base/dataset/coco_orl_stage1.py, configs/base/dataset/coco_orl_stage3.py

[2] For Algorithm-related issues,
a. Regist "ORL, Correspondence, SelectiveSearch" in mmselfsup/models/algorithms/init.py
b. Add configs/selfsup/base/models/orl.py.

[3] For Hook issues,
Regist "ORLHook" in mmselfsup/engine/hook/init.py. ORLHook is adopted in stage 1 to retrieve knn image ids of COCO train2017

[4] For Bash Scripts,
a. ORL has three stages, therefore tools/slurm_train.sh and tools/dist_train.sh are not enough.
For stage 2: Roi_generate and stage 2: Roi_pair_retrieve, add four shell scripts, i.e.
tools/dist_selective_search_single_gpu.sh, tools/slurm_selective_search_single_gpu.sh,
tools/dist_generate_correspondence_single_gpu.sh, tools/slurm_generate_correspondence_single_gpu.sh,
b. For stage2: Roi_generate and stage2: Roi_pair_retrieve, add tools/generate_correspondence.py and tools/selective_search.py

[5] For pre-train configure file
Add configure files under configs/selfsup/orl/stage1/.py, configs/selfsup/orl/stage2/.py and configs/selfsup/orl/stage3/*.py

[6] Add README.md under configs/selfsup/orl/

BC-breaking (Optional)

Does the modification introduce changes that break the backward compatibility of the downstream repositories?
If so, please describe how it breaks the compatibility and how the downstream projects should modify their code to keep compatibility with this PR.

Use cases (Optional)

If this PR introduces a new feature, it is better to list some use cases here and update the documentation.

Checklist

Before PR:

• [*] Pre-commit or other linting tools are used to fix the potential lint issues.
• Bug fixes are fully covered by unit tests, the case that causes the bug should be added in the unit tests.
• The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
• The documentation has been modified accordingly, like docstring or example tutorials.

After PR:

• If the modification has potential influence on downstream or other related projects, this PR should be tested with those projects, like MMDet or MMSeg.
• CLA has been signed and all committers have signed the CLA in this PR.

[zhaozh10]

I have tried my best to transplant ORL to mmselfsup 1.x. Unfortunately, the pre-training of ORL is multi-stage and it needs thousands of epochs on COCO train2017 dataset, hence the provided computation resource, i.e. the Beijing cloud, is insufficient. As a consequence, I hardly have the time and resources to reproduce downstream task results before DDL (1.16)

[zhaozh10]

update downstream task results (VOC07) in orl/README.md