-
Notifications
You must be signed in to change notification settings - Fork 432
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[Feature] add ORL #668
base: dev-1.x
Are you sure you want to change the base?
[Feature] add ORL #668
Conversation
add orl stage1
keep my local branch in line with the lastest branch
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) |
update downstream task results (VOC07) in orl/README.md |
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:
After PR: