Merge branch 'master' of github.com:open-mmlab/OpenSelfSup
xiaohangzhan
commit
4b74c38148
33
README.md
33
README.md
|
|
@ -20,15 +20,14 @@ Below is the relations among Unsupervised Learning, Self-Supervised Learning and
|
|||
|
||||
- **All methods in one repository**
|
||||
|
||||
* For comprehensive comparison in all benchmarks, refer to [MODEL_ZOO.md](docs/MODEL_ZOO.md).
|
||||
|
||||
<table><thead><tr><th>Method</th><th>VOC07 SVM (best layer)</th><th>ImageNet (best layer)</th></tr></thead><tbody><tr><td><a href="https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py" target="_blank" rel="noopener noreferrer">ImageNet</a></td><td>87.17</td><td>76.17</td></tr><tr><td>Random</td><td>30.22</td><td>13.70</td></tr><tr><td><a href="https://www.cv-foundation.org/openaccess/content_iccv_2015/papers/Doersch_Unsupervised_Visual_Representation_ICCV_2015_paper.pdf" target="_blank" rel="noopener noreferrer">Relative-Loc</a></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/1803.07728" target="_blank" rel="noopener noreferrer">Rotation-Pred</a></td><td>67.38</td><td>54.99</td></tr><tr><td><a href="https://arxiv.org/abs/1807.05520" target="_blank" rel="noopener noreferrer">DeepCluster</a></td><td>74.26</td><td></td></tr><tr><td><a href="https://arxiv.org/abs/1805.01978" target="_blank" rel="noopener noreferrer">NPID</a></td><td>74.50</td><td>56.61</td></tr><tr><td><a href="http://openaccess.thecvf.com/content_CVPR_2020/papers/Zhan_Online_Deep_Clustering_for_Unsupervised_Representation_Learning_CVPR_2020_paper.pdf" target="_blank" rel="noopener noreferrer">ODC</a></td><td>78.42</td><td>57.6</td></tr><tr><td><a href="https://arxiv.org/abs/1911.05722" target="_blank" rel="noopener noreferrer">MoCo</a></td><td>79.18</td><td>60.60</td></tr><tr><td><a href="https://arxiv.org/abs/2003.04297" target="_blank" rel="noopener noreferrer">MoCo v2</a></td><td>84.05</td><td>66.72</td></tr><tr><td><a href="https://arxiv.org/abs/2002.05709" target="_blank" rel="noopener noreferrer">SimCLR</a></td><td>78.95</td><td></td></tr><tr><td><a href="https://arxiv.org/abs/2006.07733" target="_blank" rel="noopener noreferrer">BYOL</a></td><td></td><td></td></tr></tbody></table>
|
||||
<em>For comprehensive comparison in all benchmarks, refer to [MODEL_ZOO.md](docs/MODEL_ZOO.md).</em>
|
||||
<table><thead><tr><th>Method</th><th>VOC07 SVM (best layer)</th><th>ImageNet (best layer)</th></tr></thead><tbody><tr><td><a href="https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py" target="_blank" rel="noopener noreferrer">ImageNet</a></td><td>87.17</td><td>76.17</td></tr><tr><td>Random</td><td>30.22</td><td>13.70</td></tr><tr><td><a href="https://www.cv-foundation.org/openaccess/content_iccv_2015/papers/Doersch_Unsupervised_Visual_Representation_ICCV_2015_paper.pdf" target="_blank" rel="noopener noreferrer">Relative-Loc</a></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/1803.07728" target="_blank" rel="noopener noreferrer">Rotation-Pred</a></td><td>67.38</td><td>54.99</td></tr><tr><td><a href="https://arxiv.org/abs/1807.05520" target="_blank" rel="noopener noreferrer">DeepCluster</a></td><td>74.26</td><td></td></tr><tr><td><a href="https://arxiv.org/abs/1805.01978" target="_blank" rel="noopener noreferrer">NPID</a></td><td>74.50</td><td>56.61</td></tr><tr><td><a href="http://openaccess.thecvf.com/content_CVPR_2020/papers/Zhan_Online_Deep_Clustering_for_Unsupervised_Representation_Learning_CVPR_2020_paper.pdf" target="_blank" rel="noopener noreferrer">ODC</a></td><td>78.42</td><td>57.6</td></tr><tr><td><a href="https://arxiv.org/abs/1911.05722" target="_blank" rel="noopener noreferrer">MoCo</a></td><td>79.18</td><td>60.60</td></tr><tr><td><a href="https://arxiv.org/abs/2003.04297" target="_blank" rel="noopener noreferrer">MoCo v2</a></td><td>84.05</td><td>66.72</td></tr><tr><td><a href="https://arxiv.org/abs/2002.05709" target="_blank" rel="noopener noreferrer">SimCLR</a></td><td>78.95</td><td></td></tr><tr><td><a href="https://arxiv.org/abs/2006.07733" target="_blank" rel="noopener noreferrer">BYOL</a></td><td></td><td></td></tr></tbody></table>
|
||||
|
||||
- **Flexibility & Extensibility**
|
||||
|
||||
`OpenSelfSup` follows a similar code architecture of MMDetection while is even more flexible than MMDetection, since OpenSelfSup integrates various self-supervised tasks including classification, joint clustering and feature learning, contrastive learning, tasks with a memory bank, etc.
|
||||
`OpenSelfSup` follows a similar code architecture of MMDetection while is even more flexible than MMDetection, since OpenSelfSup integrates various self-supervised tasks including classification, joint clustering and feature learning, contrastive learning, tasks with a memory bank, etc.
|
||||
|
||||
For existing methods in this repo, you only need to modify config files to adjust hyper-parameters. It is also simple to design your own methods, please refer to [GETTING_STARTED.md](docs/GETTING_STARTED.md).
|
||||
For existing methods in this repo, you only need to modify config files to adjust hyper-parameters. It is also simple to design your own methods, please refer to [GETTING_STARTED.md](docs/GETTING_STARTED.md).
|
||||
|
||||
- **Efficiency**
|
||||
|
||||
|
|
@ -38,21 +37,23 @@ For existing methods in this repo, you only need to modify config files to adjus
|
|||
|
||||
We standardize the benchmarks including logistic regression, SVM / Low-shot SVM from linearly probed features, semi-supervised classification, and object detection. Below are the setting of these benchmarks.
|
||||
|
||||
| Benchmarks | Setting | Remarks |
|
||||
|----------------------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------|-------------------------------------------------|
|
||||
| ImageNet Linear Classification (Multi) | [goyal2019scaling](http://openaccess.thecvf.com/content_ICCV_2019/papers/Goyal_Scaling_and_Benchmarking_Self-Supervised_Visual_Representation_Learning_ICCV_2019_paper.pdf) | Evaluate different layers. |
|
||||
| ImageNet Linear Classification (Last) | [MoCo](http://openaccess.thecvf.com/content_CVPR_2020/papers/He_Momentum_Contrast_for_Unsupervised_Visual_Representation_Learning_CVPR_2020_paper.pdf) | Evaluate the last layer after global pooling. |
|
||||
| Places205 Linear Classification | [goyal2019scaling](http://openaccess.thecvf.com/content_ICCV_2019/papers/Goyal_Scaling_and_Benchmarking_Self-Supervised_Visual_Representation_Learning_ICCV_2019_paper.pdf) | Evaluate different layers. |
|
||||
| ImageNet Semi-Sup Classification |
|
||||
| PASCAL VOC07 SVM | [goyal2019scaling](http://openaccess.thecvf.com/content_ICCV_2019/papers/Goyal_Scaling_and_Benchmarking_Self-Supervised_Visual_Representation_Learning_ICCV_2019_paper.pdf) | Costs="1.0,10.0,100.0" to save evaluation time w/o change of results. |
|
||||
| PASCAL VOC07 Low-shot SVM | [goyal2019scaling](http://openaccess.thecvf.com/content_ICCV_2019/papers/Goyal_Scaling_and_Benchmarking_Self-Supervised_Visual_Representation_Learning_ICCV_2019_paper.pdf) | Costs="1.0,10.0,100.0" to save evaluation time w/o change of results. |
|
||||
| PASCAL VOC07+12 Object Detection | [MoCo](http://openaccess.thecvf.com/content_CVPR_2020/papers/He_Momentum_Contrast_for_Unsupervised_Visual_Representation_Learning_CVPR_2020_paper.pdf) | |
|
||||
| COCO17 Object Detection | [MoCo](http://openaccess.thecvf.com/content_CVPR_2020/papers/He_Momentum_Contrast_for_Unsupervised_Visual_Representation_Learning_CVPR_2020_paper.pdf) | |
|
||||
| Benchmarks | Setting | Remarks |
|
||||
|----------------------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------|-------------------------------------------------|
|
||||
| ImageNet Linear Classification (Multi) | [goyal2019scaling](http://openaccess.thecvf.com/content_ICCV_2019/papers/Goyal_Scaling_and_Benchmarking_Self-Supervised_Visual_Representation_Learning_ICCV_2019_paper.pdf) | Evaluate different layers. |
|
||||
| ImageNet Linear Classification (Last) | [MoCo](http://openaccess.thecvf.com/content_CVPR_2020/papers/He_Momentum_Contrast_for_Unsupervised_Visual_Representation_Learning_CVPR_2020_paper.pdf) | Evaluate the last layer after global pooling. |
|
||||
| Places205 Linear Classification | [goyal2019scaling](http://openaccess.thecvf.com/content_ICCV_2019/papers/Goyal_Scaling_and_Benchmarking_Self-Supervised_Visual_Representation_Learning_ICCV_2019_paper.pdf) | Evaluate different layers. |
|
||||
| ImageNet Semi-Sup Classification |
|
||||
| PASCAL VOC07 SVM | [goyal2019scaling](http://openaccess.thecvf.com/content_ICCV_2019/papers/Goyal_Scaling_and_Benchmarking_Self-Supervised_Visual_Representation_Learning_ICCV_2019_paper.pdf) | Costs="1.0,10.0,100.0" to save evaluation time w/o change of results. |
|
||||
| PASCAL VOC07 Low-shot SVM | [goyal2019scaling](http://openaccess.thecvf.com/content_ICCV_2019/papers/Goyal_Scaling_and_Benchmarking_Self-Supervised_Visual_Representation_Learning_ICCV_2019_paper.pdf) | Costs="1.0,10.0,100.0" to save evaluation time w/o change of results. |
|
||||
| PASCAL VOC07+12 Object Detection | [MoCo](http://openaccess.thecvf.com/content_CVPR_2020/papers/He_Momentum_Contrast_for_Unsupervised_Visual_Representation_Learning_CVPR_2020_paper.pdf) | |
|
||||
| COCO17 Object Detection | [MoCo](http://openaccess.thecvf.com/content_CVPR_2020/papers/He_Momentum_Contrast_for_Unsupervised_Visual_Representation_Learning_CVPR_2020_paper.pdf) | |
|
||||
|
||||
## Change Log
|
||||
|
||||
Please refer to [CHANGELOG.md](docs/CHANGELOG.md) for details and release history.
|
||||
|
||||
[2020-06-26] `OpenSelfSup` v0.2.0 is released with benchmark results and support of new features.
|
||||
|
||||
[2020-06-16] `OpenSelfSup` v0.1.0 is released.
|
||||
|
||||
## Installation
|
||||
|
|
@ -94,4 +95,4 @@ fair_self_supervision_benchmark](https://github.com/facebookresearch/fair_self_s
|
|||
|
||||
## Contact
|
||||
|
||||
This repo is currently maintained by Xiaohang Zhan ([@XiaohangZhan](http://github.com/XiaohangZhan)).
|
||||
This repo is currently maintained by Xiaohang Zhan ([@XiaohangZhan](http://github.com/XiaohangZhan)) and Jiahao Xie ([@Jiahao000](https://github.com/Jiahao000)).
|
||||
|
|
|
|||
|
|
@ -1,4 +1,4 @@
|
|||
# Getting Started
|
||||
se# Getting Started
|
||||
|
||||
This page provides basic tutorials about the usage of OpenSelfSup.
|
||||
For installation instructions, please see [INSTALL.md](INSTALL.md).
|
||||
|
|
@ -85,7 +85,7 @@ Augments:
|
|||
Working directories:
|
||||
The features, logs and intermediate files generated are saved in `$SVM_WORK_DIR/` as follows:
|
||||
- `dist_test_svm_epoch.sh`: `SVM_WORK_DIR=$WORK_DIR/` (The same as that mentioned in `Train with single/multiple GPUs` above.) Hence, the files will be overridden to save space when evaluating with a new `$EPOCH`.
|
||||
- `dist_test_svm_pretrain.sh`: `SVM_WORK_DIR=$WORK_DIR/$PRETRAIN_NAME/`, e.g., if `PRETRAIN=pretrains/odc_v1.pth`, then `PRETRAIN_NAME=odc_v1.pth`; if `PRETRAIN=random`, then `PRETRAIN_NAME=random`.
|
||||
- `dist_test_svm_pretrain.sh`: `SVM_WORK_DIR=$WORK_DIR/$PRETRAIN_NAME/`, e.g., if `PRETRAIN=pretrains/odc_r50_v1-5af5dd0c.pth`, then `PRETRAIN_NAME=odc_r50_v1-5af5dd0c.pth`; if `PRETRAIN=random`, then `PRETRAIN_NAME=random`.
|
||||
The evaluation records are saved in `$SVM_WORK_DIR/logs/eval_svm.log`
|
||||
|
||||
### ImageNet / Places205 Linear Classification
|
||||
|
|
@ -96,7 +96,7 @@ python tools/extract_backbone_weights.py ${CHECKPOINT} ${WEIGHT_FILE}
|
|||
```
|
||||
Arguments:
|
||||
- `CHECKPOINTS`: the checkpoint file of a selfsup method named as `epoch_*.pth`.
|
||||
- `WEIGHT_FILE`: the output backbone weights file, e.g., `pretrains/moco_v1_epoch200.pth`.
|
||||
- `WEIGHT_FILE`: the output backbone weights file, e.g., `pretrains/moco_r50_v1-4ad89b5c.pth`.
|
||||
|
||||
**Next**, train and test linear classification:
|
||||
```shell
|
||||
|
|
@ -112,13 +112,13 @@ Augments:
|
|||
- `--deterministic`: Switch on "deterministic" mode which slows down training but the results are reproducible.
|
||||
|
||||
Working directories:
|
||||
Where are the checkpoints and logs? E.g., if you use `configs/benchmarks/linear_classification/imagenet/r50_multihead.py` to evaluate `pretrains/moco_v1_epoch200.pth`, then the working directories for this evalution is `work_dirs/benchmarks/linear_classification/imagenet/r50_multihead/moco_v1_epoch200.pth/`.
|
||||
Where are the checkpoints and logs? E.g., if you use `configs/benchmarks/linear_classification/imagenet/r50_multihead.py` to evaluate `pretrains/moco_r50_v1-4ad89b5c.pth`, then the working directories for this evalution is `work_dirs/benchmarks/linear_classification/imagenet/r50_multihead/moco_r50_v1-4ad89b5c.pth/`.
|
||||
|
||||
### ImageNet Semi-Supervised Classification
|
||||
|
||||
```shell
|
||||
# train
|
||||
bash benchmarks/dist_train_linear.sh ${CONFIG_FILE} ${WEIGHT_FILE} [optional arguments]
|
||||
bash benchmarks/dist_train_semi.sh ${CONFIG_FILE} ${WEIGHT_FILE} [optional arguments]
|
||||
# test (unnecessary if have validation in training)
|
||||
bash tools/dist_test.sh ${CONFIG_FILE} ${GPUS} ${CHECKPOINT}
|
||||
```
|
||||
|
|
|
|||
|
|
@ -2,17 +2,118 @@
|
|||
|
||||
## Pre-trained model download links
|
||||
|
||||
<table><thead><tr><th>Method</th><th>Config</th><th>Remarks</th><th>Download link</th></tr></thead><tbody><tr><td><a href="https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py" target="_blank" rel="noopener noreferrer">ImageNet</a></td><td>-</td><td>torchvision</td><td><a href="https://drive.google.com/file/d/11xA3TOcbD0qOrwpBfYonEDeseE1wMfBh/view?usp=sharing" target="_blank" rel="noopener noreferrer">imagenet_r50-21352794.pth</a></td></tr><tr><td>Random</td><td>-</td><td>kaiming</td><td><a href="https://drive.google.com/file/d/1UaFTjd6sbKkZEE-f58Zv30bnx7C1qJBb/view?usp=sharing" target="_blank" rel="noopener noreferrer">random_r50-5d0fa71b.pth</a></td></tr><tr><td><a href="https://www.cv-foundation.org/openaccess/content_iccv_2015/papers/Doersch_Unsupervised_Visual_Representation_ICCV_2015_paper.pdf" target="_blank" rel="noopener noreferrer">Relative-Loc</a></td><td>selfsup/relative_loc/r50.py</td><td>default</td><td></td></tr><tr><td><a href="https://arxiv.org/abs/1803.07728" target="_blank" rel="noopener noreferrer">Rotation-Pred</a></td><td>selfsup/rotation_pred/r50.py</td><td>default</td><td><a href="https://drive.google.com/file/d/1t3oClmIvQ0p8RZ0V5yvQFltzjqBO823Y/view?usp=sharing" target="_blank" rel="noopener noreferrer">rotation_r50-cfab8ebb.pth</a></td></tr><tr><td><a href="https://arxiv.org/abs/1807.05520" target="_blank" rel="noopener noreferrer">DeepCluster</a></td><td>selfsup/deepcluster/r50.py</td><td>default</td><td><a href="https://drive.google.com/file/d/1GxgP7pI18JtFxDIC0hnHOanvUYajoLlg/view?usp=sharing" target="_blank" rel="noopener noreferrer">deepcluster_r50-bb8681e2.pth</a></td></tr><tr><td><a href="https://arxiv.org/abs/1805.01978" target="_blank" rel="noopener noreferrer">NPID</a></td><td>selfsup/npid/r50.py</td><td>default</td><td><a href="https://drive.google.com/file/d/1sm6I3Y5XnCWdbmeLSF4YupUtPe5nRQMI/view?usp=sharing" target="_blank" rel="noopener noreferrer">npid_r50-dec3df0c.pth</a></td></tr><tr><td><a href="http://openaccess.thecvf.com/content_CVPR_2020/papers/Zhan_Online_Deep_Clustering_for_Unsupervised_Representation_Learning_CVPR_2020_paper.pdf" target="_blank" rel="noopener noreferrer">ODC</a></td><td>selfsup/odc/r50_v1.py</td><td>default</td><td><a href="https://drive.google.com/file/d/1EdhJeZAyMsD_pEW7uMhLzos5xZLdariN/view?usp=sharing" target="_blank" rel="noopener noreferrer">odc_r50_v1-5af5dd0c.pth</a></td></tr><tr><td><a href="https://arxiv.org/abs/1911.05722" target="_blank" rel="noopener noreferrer">MoCo</a></td><td>selfsup/moco/r50_v1.py</td><td>default</td><td><a href="https://drive.google.com/file/d/1ANXfnoT8yBQQBBqR_kQLQorK20l65KMy/view?usp=sharing" target="_blank" rel="noopener noreferrer">moco_r50_v1-4ad89b5c.pth</a></td></tr><tr><td><a href="https://arxiv.org/abs/2003.04297" target="_blank" rel="noopener noreferrer">MoCo v2</a></td><td>selfsup/moco/r50_v2.py</td><td>default</td><td><a href="https://drive.google.com/file/d/1Cc5qMjPkKW6WeM4ic9Tq-IBxswyJhMnF/view?usp=sharing" target="_blank" rel="noopener noreferrer">moco_r50_v2-58f10cfe.pth</a></td></tr><tr><td></td><td>selfsup/moco/r50_v2_simclr_neck.py</td><td>-> SimCLR neck<br></td><td><a href="https://drive.google.com/file/d/1PnZmCVmFwBv7ZnqMgNYj5DvmbPGM5rCx/view?usp=sharing" target="_blank" rel="noopener noreferrer">moco_r50_v2_simclr_neck-70379356.pth</a></td></tr><tr><td><a href="https://arxiv.org/abs/2002.05709" target="_blank" rel="noopener noreferrer">SimCLR</a></td><td>selfsup/simclr/r50_bs256_ep200.py</td><td>default</td><td><a href="https://drive.google.com/file/d/1aZ43nSdivdNxHbM9DKVoZYVhZ8TNnmPp/view?usp=sharing" target="_blank" rel="noopener noreferrer">simclr_r50_bs256_ep200-4577e9a6.pth</a></td></tr><tr><td></td><td>selfsup/simclr/r50_bs256_ep200_mocov2_neck.py</td><td>-> MoCo v2 neck</td><td><a href="https://drive.google.com/file/d/1AXpSKqgWfnj6jCgN65BXSTCKFfuIVELa/view?usp=sharing" target="_blank" rel="noopener noreferrer">simclr_r50_bs256_ep200_mocov2_neck-0d6e5ff2.pth</a></td></tr><tr><td><a href="https://arxiv.org/abs/2006.07733" target="_blank" rel="noopener noreferrer">BYOL</a></td><td>selfsup/byol/r50.py</td><td>default</td><td></td></tr></tbody></table>
|
||||
<table>
|
||||
<thead>
|
||||
<tr>
|
||||
<th>Method</th>
|
||||
<th>Config</th>
|
||||
<th>Remarks</th>
|
||||
<th>Download link</th>
|
||||
</tr>
|
||||
</thead>
|
||||
<tbody>
|
||||
<tr>
|
||||
<td><a href="https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py" target="_blank" rel="noopener noreferrer">ImageNet</a></td>
|
||||
<td>-</td>
|
||||
<td>torchvision</td>
|
||||
<td><a href="https://drive.google.com/file/d/11xA3TOcbD0qOrwpBfYonEDeseE1wMfBh/view?usp=sharing" target="_blank" rel="noopener noreferrer">imagenet_r50-21352794.pth</a></td>
|
||||
</tr>
|
||||
<tr>
|
||||
<td>Random</td>
|
||||
<td>-</td>
|
||||
<td>kaiming</td>
|
||||
<td><a href="https://drive.google.com/file/d/1UaFTjd6sbKkZEE-f58Zv30bnx7C1qJBb/view?usp=sharing" target="_blank" rel="noopener noreferrer">random_r50-5d0fa71b.pth</a></td>
|
||||
</tr>
|
||||
<tr>
|
||||
<td><a href="https://www.cv-foundation.org/openaccess/content_iccv_2015/papers/Doersch_Unsupervised_Visual_Representation_ICCV_2015_paper.pdf" target="_blank" rel="noopener noreferrer">Relative-Loc</a></td>
|
||||
<td>selfsup/relative_loc/r50.py</td>
|
||||
<td>default</td>
|
||||
<td></td>
|
||||
</tr>
|
||||
<tr>
|
||||
<td><a href="https://arxiv.org/abs/1803.07728" target="_blank" rel="noopener noreferrer">Rotation-Pred</a></td>
|
||||
<td>selfsup/rotation_pred/r50.py</td>
|
||||
<td>default</td>
|
||||
<td><a href="https://drive.google.com/file/d/1t3oClmIvQ0p8RZ0V5yvQFltzjqBO823Y/view?usp=sharing" target="_blank" rel="noopener noreferrer">rotation_r50-cfab8ebb.pth</a></td>
|
||||
</tr>
|
||||
<tr>
|
||||
<td><a href="https://arxiv.org/abs/1807.05520" target="_blank" rel="noopener noreferrer">DeepCluster</a></td>
|
||||
<td>selfsup/deepcluster/r50.py</td>
|
||||
<td>default</td>
|
||||
<td><a href="https://drive.google.com/file/d/1GxgP7pI18JtFxDIC0hnHOanvUYajoLlg/view?usp=sharing" target="_blank" rel="noopener noreferrer">deepcluster_r50-bb8681e2.pth</a></td>
|
||||
</tr>
|
||||
<tr>
|
||||
<td><a href="https://arxiv.org/abs/1805.01978" target="_blank" rel="noopener noreferrer">NPID</a></td>
|
||||
<td>selfsup/npid/r50.py</td>
|
||||
<td>default</td>
|
||||
<td><a href="https://drive.google.com/file/d/1sm6I3Y5XnCWdbmeLSF4YupUtPe5nRQMI/view?usp=sharing" target="_blank" rel="noopener noreferrer">npid_r50-dec3df0c.pth</a></td>
|
||||
</tr>
|
||||
<tr>
|
||||
<td></td>
|
||||
<td>selfsup/npid/r50_ensure_neg.py</td>
|
||||
<td>default</td>
|
||||
<td><a href="https://drive.google.com/file/d/1FldDrb6kzF3CZ7737mwCXVI6HE2aCSaF/view?usp=sharing" target="_blank" rel="noopener noreferrer">npid_r50_ensure_neg-ce09b7ae.pth</a></td>
|
||||
</tr>
|
||||
<tr>
|
||||
<td><a href="http://openaccess.thecvf.com/content_CVPR_2020/papers/Zhan_Online_Deep_Clustering_for_Unsupervised_Representation_Learning_CVPR_2020_paper.pdf" target="_blank" rel="noopener noreferrer">ODC</a></td>
|
||||
<td>selfsup/odc/r50_v1.py</td>
|
||||
<td>default</td>
|
||||
<td><a href="https://drive.google.com/file/d/1EdhJeZAyMsD_pEW7uMhLzos5xZLdariN/view?usp=sharing" target="_blank" rel="noopener noreferrer">odc_r50_v1-5af5dd0c.pth</a></td>
|
||||
</tr>
|
||||
<tr>
|
||||
<td><a href="https://arxiv.org/abs/1911.05722" target="_blank" rel="noopener noreferrer">MoCo</a></td>
|
||||
<td>selfsup/moco/r50_v1.py</td>
|
||||
<td>default</td>
|
||||
<td><a href="https://drive.google.com/file/d/1ANXfnoT8yBQQBBqR_kQLQorK20l65KMy/view?usp=sharing" target="_blank" rel="noopener noreferrer">moco_r50_v1-4ad89b5c.pth</a></td>
|
||||
</tr>
|
||||
<tr>
|
||||
<td><a href="https://arxiv.org/abs/2003.04297" target="_blank" rel="noopener noreferrer">MoCo v2</a></td>
|
||||
<td>selfsup/moco/r50_v2.py</td>
|
||||
<td>default</td>
|
||||
<td><a href="https://drive.google.com/file/d/1Cc5qMjPkKW6WeM4ic9Tq-IBxswyJhMnF/view?usp=sharing" target="_blank" rel="noopener noreferrer">moco_r50_v2-58f10cfe.pth</a></td>
|
||||
</tr>
|
||||
<tr>
|
||||
<td></td>
|
||||
<td>selfsup/moco/r50_v2_simclr_neck.py</td>
|
||||
<td>-> SimCLR neck<br></td>
|
||||
<td><a href="https://drive.google.com/file/d/1PnZmCVmFwBv7ZnqMgNYj5DvmbPGM5rCx/view?usp=sharing" target="_blank" rel="noopener noreferrer">moco_r50_v2_simclr_neck-70379356.pth</a></td>
|
||||
</tr>
|
||||
<tr>
|
||||
<td><a href="https://arxiv.org/abs/2002.05709" target="_blank" rel="noopener noreferrer">SimCLR</a></td>
|
||||
<td>selfsup/simclr/r50_bs256_ep200.py</td>
|
||||
<td>default</td>
|
||||
<td><a href="https://drive.google.com/file/d/1aZ43nSdivdNxHbM9DKVoZYVhZ8TNnmPp/view?usp=sharing" target="_blank" rel="noopener noreferrer">simclr_r50_bs256_ep200-4577e9a6.pth</a></td>
|
||||
</tr>
|
||||
<tr>
|
||||
<td></td>
|
||||
<td>selfsup/simclr/r50_bs256_ep200_mocov2_neck.py</td>
|
||||
<td>-> MoCo v2 neck</td>
|
||||
<td><a href="https://drive.google.com/file/d/1AXpSKqgWfnj6jCgN65BXSTCKFfuIVELa/view?usp=sharing" target="_blank" rel="noopener noreferrer">simclr_r50_bs256_ep200_mocov2_neck-0d6e5ff2.pth</a></td>
|
||||
</tr>
|
||||
<tr>
|
||||
<td><a href="https://arxiv.org/abs/2006.07733" target="_blank" rel="noopener noreferrer">BYOL</a></td>
|
||||
<td>selfsup/byol/r50.py</td>
|
||||
<td>default</td>
|
||||
<td></td>
|
||||
</tr>
|
||||
</tbody>
|
||||
</table>
|
||||
|
||||
## Benchmarks
|
||||
|
||||
### VOC07 SVM & SVM Low-shot
|
||||
|
||||
<table><thead><tr><th rowspan="2">Method</th><th rowspan="2">Config</th><th rowspan="2">Remarks</th><th rowspan="2">Best layer</th><th rowspan="2">VOC07 SVM</th><th colspan="8">VOC07 SVM Low-shot</th></tr><tr><td>1</td><td>2</td><td>4</td><td>8</td><td>16</td><td>32</td><td>64</td><td>96</td></tr></thead><tbody><tr><td><a href="https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py" target="_blank" rel="noopener noreferrer">ImageNet</a></td><td>-</td><td>torchvision</td><td>feat5</td><td>87.17</td><td>52.99</td><td>63.55</td><td>73.7</td><td>78.79</td><td>81.76</td><td>83.75</td><td>85.18</td><td>85.97</td></tr><tr><td>Random</td><td>-</td><td>kaiming</td><td>feat2</td><td>30.22</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="https://www.cv-foundation.org/openaccess/content_iccv_2015/papers/Doersch_Unsupervised_Visual_Representation_ICCV_2015_paper.pdf" target="_blank" rel="noopener noreferrer">Relative-Loc</a></td><td></td><td></td><td>feat5</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/1803.07728" target="_blank" rel="noopener noreferrer">Rotation-Pred</a></td><td>selfsup/rotation_pred/r50.py</td><td>default</td><td>feat4</td><td>67.38</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/1807.05520" target="_blank" rel="noopener noreferrer">DeepCluster</a></td><td>selfsup/deepcluster/r50.py</td><td>default</td><td>feat5</td><td>74.26</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/1805.01978" target="_blank" rel="noopener noreferrer">NPID</a></td><td>selfsup/npid/r50.py</td><td>default</td><td>feat5</td><td>74.50</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="http://openaccess.thecvf.com/content_CVPR_2020/papers/Zhan_Online_Deep_Clustering_for_Unsupervised_Representation_Learning_CVPR_2020_paper.pdf" target="_blank" rel="noopener noreferrer">ODC</a></td><td>selfsup/odc/r50_v1.py</td><td>default</td><td>feat5</td><td>78.42</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/1911.05722" target="_blank" rel="noopener noreferrer">MoCo</a></td><td>selfsup/moco/r50_v1.py</td><td>default</td><td>feat5</td><td>79.18</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/2003.04297" target="_blank" rel="noopener noreferrer">MoCo v2</a></td><td>selfsup/moco/r50_v2.py</td><td>default</td><td>feat5</td><td>84.05</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td></td><td>selfsup/moco/r50_v2_simclr_neck.py</td><td>-> SimCLR neck<br></td><td>feat5</td><td>84.00</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/2002.05709" target="_blank" rel="noopener noreferrer">SimCLR</a></td><td>selfsup/simclr/r50_bs256_ep200.py</td><td>default</td><td>feat5</td><td>78.95</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td></td><td>selfsup/simclr/r50_bs256_ep200_mocov2_neck.py</td><td>-> MoCo v2 neck</td><td>feat5</td><td>77.65</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/2006.07733" target="_blank" rel="noopener noreferrer">BYOL</a></td><td>selfsup/byol/r50.py</td><td>default</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr></tbody></table>
|
||||
<table><thead><tr><th rowspan="2">Method</th><th rowspan="2">Config</th><th rowspan="2">Remarks</th><th rowspan="2">Best layer</th><th rowspan="2">VOC07 SVM</th><th colspan="8">VOC07 SVM Low-shot</th></tr><tr><td>1</td><td>2</td><td>4</td><td>8</td><td>16</td><td>32</td><td>64</td><td>96</td></tr></thead><tbody><tr><td><a href="https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py" target="_blank" rel="noopener noreferrer">ImageNet</a></td><td>-</td><td>torchvision</td><td>feat5</td><td>87.17</td><td>52.99</td><td>63.55</td><td>73.7</td><td>78.79</td><td>81.76</td><td>83.75</td><td>85.18</td><td>85.97</td></tr><tr><td>Random</td><td>-</td><td>kaiming</td><td>feat2</td><td>30.22</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="https://www.cv-foundation.org/openaccess/content_iccv_2015/papers/Doersch_Unsupervised_Visual_Representation_ICCV_2015_paper.pdf" target="_blank" rel="noopener noreferrer">Relative-Loc</a></td><td></td><td></td><td>feat5</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/1803.07728" target="_blank" rel="noopener noreferrer">Rotation-Pred</a></td><td>selfsup/rotation_pred/r50.py</td><td>default</td><td>feat4</td><td>67.38</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/1807.05520" target="_blank" rel="noopener noreferrer">DeepCluster</a></td><td>selfsup/deepcluster/r50.py</td><td>default</td><td>feat5</td><td>74.26</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/1805.01978" target="_blank" rel="noopener noreferrer">NPID</a></td><td>selfsup/npid/r50.py</td><td>default</td><td>feat5</td><td>74.50</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td></td><td>selfsup/npid/r50_ensure_neg.py</td><td>ensure_neg=True</td><td>feat5</td><td>75.70</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="http://openaccess.thecvf.com/content_CVPR_2020/papers/Zhan_Online_Deep_Clustering_for_Unsupervised_Representation_Learning_CVPR_2020_paper.pdf" target="_blank" rel="noopener noreferrer">ODC</a></td><td>selfsup/odc/r50_v1.py</td><td>default</td><td>feat5</td><td>78.42</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/1911.05722" target="_blank" rel="noopener noreferrer">MoCo</a></td><td>selfsup/moco/r50_v1.py</td><td>default</td><td>feat5</td><td>79.18</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/2003.04297" target="_blank" rel="noopener noreferrer">MoCo v2</a></td><td>selfsup/moco/r50_v2.py</td><td>default</td><td>feat5</td><td>84.05</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td></td><td>selfsup/moco/r50_v2_simclr_neck.py</td><td>-> SimCLR neck<br></td><td>feat5</td><td>84.00</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/2002.05709" target="_blank" rel="noopener noreferrer">SimCLR</a></td><td>selfsup/simclr/r50_bs256_ep200.py</td><td>default</td><td>feat5</td><td>78.95</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td></td><td>selfsup/simclr/r50_bs256_ep200_mocov2_neck.py</td><td>-> MoCo v2 neck</td><td>feat5</td><td>77.65</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/2006.07733" target="_blank" rel="noopener noreferrer">BYOL</a></td><td>selfsup/byol/r50.py</td><td>default</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr></tbody></table>
|
||||
|
||||
### ImageNet Linear Classification
|
||||
|
||||
<table><thead><tr><th rowspan="2">Method</th><th rowspan="2">Config</th><th rowspan="2">Remarks</th><th colspan="5">ImageNet (Multi)</th><th>ImageNet (Last)</th></tr><tr><td>feat1</td><td>feat2</td><td>feat3</td><td>feat4</td><td>feat5</td><td>avgpool</td></tr></thead><tbody><tr><td><a href="https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py" target="_blank" rel="noopener noreferrer">ImageNet</a></td><td>-</td><td>torchvision</td><td>15.18</td><td>33.96</td><td>47.86</td><td>67.56</td><td>76.17</td><td>74.12</td></tr><tr><td>Random</td><td>-</td><td>kaiming</td><td></td><td></td><td></td><td></td><td></td><td>4.35</td></tr><tr><td><a href="https://www.cv-foundation.org/openaccess/content_iccv_2015/papers/Doersch_Unsupervised_Visual_Representation_ICCV_2015_paper.pdf" target="_blank" rel="noopener noreferrer">Relative-Loc</a></td><td>selfsup/relative_loc/r50.py</td><td>default</td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/1803.07728" target="_blank" rel="noopener noreferrer">Rotation-Pred</a></td><td>selfsup/rotation_pred/r50.py</td><td>default</td><td>12.89</td><td>34.30</td><td>44.91</td><td>54.99</td><td>49.09</td><td></td></tr><tr><td><a href="https://arxiv.org/abs/1807.05520" target="_blank" rel="noopener noreferrer">DeepCluster</a></td><td>selfsup/deepcluster/r50.py</td><td>default</td><td></td><td></td><td></td><td></td><td></td><td>46.92</td></tr><tr><td><a href="https://arxiv.org/abs/1805.01978" target="_blank" rel="noopener noreferrer">NPID</a></td><td>selfsup/npid/r50.py</td><td>default</td><td>14.28</td><td>31.20</td><td>40.68</td><td>54.46</td><td>56.61</td><td></td></tr><tr><td><a href="http://openaccess.thecvf.com/content_CVPR_2020/papers/Zhan_Online_Deep_Clustering_for_Unsupervised_Representation_Learning_CVPR_2020_paper.pdf" target="_blank" rel="noopener noreferrer">ODC</a></td><td>selfsup/odc/r50_v1.py</td><td>default</td><td>14.75</td><td>31.55</td><td>42.49</td><td>55.72</td><td>57.57</td><td></td></tr><tr><td><a href="https://arxiv.org/abs/1911.05722" target="_blank" rel="noopener noreferrer">MoCo</a></td><td>selfsup/moco/r50_v1.py</td><td>default</td><td>15.32</td><td>33.08</td><td>44.68</td><td>57.27</td><td>60.60</td><td>61.02</td></tr><tr><td><a href="https://arxiv.org/abs/2003.04297" target="_blank" rel="noopener noreferrer">MoCo v2</a></td><td>selfsup/moco/r50_v2.py</td><td>default</td><td>15.35</td><td>34.57</td><td>45.81</td><td>60.96</td><td>66.72</td><td>67.02</td></tr><tr><td></td><td>selfsup/moco/r50_v2_simclr_neck.py</td><td>-> SimCLR neck<br></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/2002.05709" target="_blank" rel="noopener noreferrer">SimCLR</a></td><td>selfsup/simclr/r50_bs256_ep200.py</td><td>default</td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td></td><td>selfsup/simclr/r50_bs256_ep200_mocov2_neck.py</td><td>-> MoCo v2 neck</td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/2006.07733" target="_blank" rel="noopener noreferrer">BYOL</a></td><td>selfsup/byol/r50.py</td><td>default</td><td></td><td></td><td></td><td></td><td></td><td></td></tr></tbody></table>
|
||||
**NOte**
|
||||
* Config: `configs/benchmarks/linear_classification/imagenet/r50_multihead.py` for ImageNet (Multi) and `configs/benchmarks/linear_classification/imagenet/r50_moco.py` for ImageNet (Last).
|
||||
* For DeepCluster, use the corresponding one with `_sobel`.
|
||||
* ImageNet (Multi) evaluates features in around 9k dimensions from different layers. Top-1 result of the last epoch is reported.
|
||||
* ImageNet (Last) evaluates the last feature after global average pooling, e.g., 2048 dimensions for resnet50. The best top-1 result among all epochs is reported.
|
||||
|
||||
<table><thead><tr><th rowspan="2">Method</th><th rowspan="2">Config</th><th rowspan="2">Remarks</th><th colspan="5">ImageNet (Multi)</th><th>ImageNet (Last)</th></tr><tr><td>feat1</td><td>feat2</td><td>feat3</td><td>feat4</td><td>feat5</td><td>avgpool</td></tr></thead><tbody><tr><td><a href="https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py" target="_blank" rel="noopener noreferrer">ImageNet</a></td><td>-</td><td>torchvision</td><td>15.18</td><td>33.96</td><td>47.86</td><td>67.56</td><td>76.17</td><td>74.12</td></tr><tr><td>Random</td><td>-</td><td>kaiming</td><td></td><td></td><td></td><td></td><td></td><td>4.35</td></tr><tr><td><a href="https://www.cv-foundation.org/openaccess/content_iccv_2015/papers/Doersch_Unsupervised_Visual_Representation_ICCV_2015_paper.pdf" target="_blank" rel="noopener noreferrer">Relative-Loc</a></td><td>selfsup/relative_loc/r50.py</td><td>default</td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/1803.07728" target="_blank" rel="noopener noreferrer">Rotation-Pred</a></td><td>selfsup/rotation_pred/r50.py</td><td>default</td><td>12.89</td><td>34.30</td><td>44.91</td><td>54.99</td><td>49.09</td><td>47.01</td></tr><tr><td><a href="https://arxiv.org/abs/1807.05520" target="_blank" rel="noopener noreferrer">DeepCluster</a></td><td>selfsup/deepcluster/r50.py</td><td>default</td><td></td><td></td><td></td><td></td><td></td><td>46.92</td></tr><tr><td><a href="https://arxiv.org/abs/1805.01978" target="_blank" rel="noopener noreferrer">NPID</a></td><td>selfsup/npid/r50.py</td><td>default</td><td>14.28</td><td>31.20</td><td>40.68</td><td>54.46</td><td>56.61</td><td>56.60</td></tr><tr><td><a href="http://openaccess.thecvf.com/content_CVPR_2020/papers/Zhan_Online_Deep_Clustering_for_Unsupervised_Representation_Learning_CVPR_2020_paper.pdf" target="_blank" rel="noopener noreferrer">ODC</a></td><td>selfsup/odc/r50_v1.py</td><td>default</td><td>14.76</td><td>31.82</td><td>42.44</td><td>55.76</td><td>57.70</td><td>53.42</td></tr><tr><td><a href="https://arxiv.org/abs/1911.05722" target="_blank" rel="noopener noreferrer">MoCo</a></td><td>selfsup/moco/r50_v1.py</td><td>default</td><td>15.32</td><td>33.08</td><td>44.68</td><td>57.27</td><td>60.60</td><td>61.02</td></tr><tr><td><a href="https://arxiv.org/abs/2003.04297" target="_blank" rel="noopener noreferrer">MoCo v2</a></td><td>selfsup/moco/r50_v2.py</td><td>default</td><td>15.35</td><td>34.57</td><td>45.81</td><td>60.96</td><td>66.72</td><td>67.02</td></tr><tr><td></td><td>selfsup/moco/r50_v2_simclr_neck.py</td><td>-> SimCLR neck<br></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/2002.05709" target="_blank" rel="noopener noreferrer">SimCLR</a></td><td>selfsup/simclr/r50_bs256_ep200.py</td><td>default</td><td>17.09</td><td>31.37</td><td>41.38</td><td>54.35</td><td>61.57</td><td>60.06</td></tr><tr><td></td><td>selfsup/simclr/r50_bs256_ep200_mocov2_neck.py</td><td>-> MoCo v2 neck</td><td>16.97</td><td>31.88</td><td>41.73</td><td>54.33</td><td>59.94</td><td>58.00</td></tr><tr><td><a href="https://arxiv.org/abs/2006.07733" target="_blank" rel="noopener noreferrer">BYOL</a></td><td>selfsup/byol/r50.py</td><td>default</td><td></td><td></td><td></td><td></td><td></td><td></td></tr></tbody></table>
|
||||
|
||||
### Place Linear Classification
|
||||
|
||||
|
|
@ -22,8 +123,12 @@ Coming soon.
|
|||
|
||||
**Note**
|
||||
* In this benchmark, the necks or heads are removed and only the backbone CNN is evaluated by appending a linear classification head. All parameters are fine-tuned.
|
||||
* When training with 1% ImageNet, we find hyper-parameters especially the learning rate greatly influence the performance. Hence, we prepare a list of settings with the base learning rate from \{0.01, 0.1\} and the learning rate multiplier for the head from \{1, 10, 100\}. We choose the best performing setting for each method.
|
||||
* Config: under `configs/benchmarks/semi_classification/imagenet_1percent/` for 1% data, and `configs/benchmarks/semi_classification/imagenet_10percent/` for 10% data.
|
||||
* When training with 1% ImageNet, we find hyper-parameters especially the learning rate greatly influence the performance. Hence, we prepare a list of settings with the base learning rate from \{0.001, 0.01, 0.1\} and the learning rate multiplier for the head from \{1, 10, 100\}. We choose the best performing setting for each method.
|
||||
* Please use `--deterministic` in this benchmark.
|
||||
|
||||
<table><thead><tr><th rowspan="2">Method</th><th rowspan="2">Config</th><th rowspan="2">Remarks</th><th rowspan="2">Optimal setting for ImageNet 1%</th><th colspan="2">ImageNet 1%</th></tr><tr><td>top-1</td><td>top-5</td></tr></thead><tbody><tr><td><a href="https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py" target="_blank" rel="noopener noreferrer">ImageNet</a></td><td>-</td><td>torchvision</td><td>r50_lr0_001_head100.py</td><td>68.68</td><td>88.87</td></tr><tr><td>Random</td><td>-</td><td>kaiming</td><td>r50_lr0_01_head1.py</td><td>1.56</td><td>4.99</td></tr><tr><td><a href="https://www.cv-foundation.org/openaccess/content_iccv_2015/papers/Doersch_Unsupervised_Visual_Representation_ICCV_2015_paper.pdf" target="_blank" rel="noopener noreferrer">Relative-Loc</a></td><td>selfsup/relative_loc/r50.py</td><td>default</td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/1803.07728" target="_blank" rel="noopener noreferrer">Rotation-Pred</a></td><td>selfsup/rotation_pred/r50.py</td><td>default</td><td>r50_lr0_01_head100.py</td><td>18.98</td><td>44.05</td></tr><tr><td><a href="https://arxiv.org/abs/1807.05520" target="_blank" rel="noopener noreferrer">DeepCluster</a></td><td>selfsup/deepcluster/r50.py</td><td>default</td><td>r50_lr0_01_head1_sobel.py</td><td>33.44</td><td>58.62</td></tr><tr><td><a href="https://arxiv.org/abs/1805.01978" target="_blank" rel="noopener noreferrer">NPID</a></td><td>selfsup/npid/r50.py</td><td>default</td><td>r50_lr0_01_head100.py</td><td>27.95</td><td>54.37</td></tr><tr><td><a href="http://openaccess.thecvf.com/content_CVPR_2020/papers/Zhan_Online_Deep_Clustering_for_Unsupervised_Representation_Learning_CVPR_2020_paper.pdf" target="_blank" rel="noopener noreferrer">ODC</a></td><td>selfsup/odc/r50_v1.py</td><td>default</td><td>r50_lr0_1_head100.py</td><td>32.39</td><td>61.02</td></tr><tr><td><a href="https://arxiv.org/abs/1911.05722" target="_blank" rel="noopener noreferrer">MoCo</a></td><td>selfsup/moco/r50_v1.py</td><td>default</td><td>r50_lr0_01_head100.py</td><td>33.15</td><td>61.30</td></tr><tr><td><a href="https://arxiv.org/abs/2003.04297" target="_blank" rel="noopener noreferrer">MoCo v2</a></td><td>selfsup/moco/r50_v2.py</td><td>default</td><td>r50_lr0_01_head100.py</td><td>38.71</td><td>67.90</td></tr><tr><td></td><td>selfsup/moco/r50_v2_simclr_neck.py</td><td>-> SimCLR neck<br></td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/2002.05709" target="_blank" rel="noopener noreferrer">SimCLR</a></td><td>selfsup/simclr/r50_bs256_ep200.py</td><td>default</td><td>r50_lr0_01_head100.py</td><td>36.09</td><td>64.50</td></tr><tr><td></td><td>selfsup/simclr/r50_bs256_ep200_mocov2_neck.py</td><td>-> MoCo v2 neck</td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/2006.07733" target="_blank" rel="noopener noreferrer">BYOL</a></td><td>selfsup/byol/r50.py</td><td>default</td><td></td><td></td><td></td></tr></tbody></table>
|
||||
|
||||
<table><thead><tr><th rowspan="2">Method</th><th rowspan="2">Config</th><th rowspan="2">Remarks</th><th rowspan="2">Optimal setting for ImageNet 1%</th><th colspan="2">ImageNet 1%</th></tr><tr><td>top-1</td><td>top-5</td></tr></thead><tbody><tr><td><a href="https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py" target="_blank" rel="noopener noreferrer">ImageNet</a></td><td>-</td><td>torchvision</td><td>r50_lr0_01_head1.py</td><td>63.10</td><td>85.73</td></tr><tr><td>Random</td><td>-</td><td>kaiming</td><td>r50_lr0_01_head1.py</td><td>1.56</td><td>4.99</td></tr><tr><td><a href="https://www.cv-foundation.org/openaccess/content_iccv_2015/papers/Doersch_Unsupervised_Visual_Representation_ICCV_2015_paper.pdf" target="_blank" rel="noopener noreferrer">Relative-Loc</a></td><td>selfsup/relative_loc/r50.py</td><td>default</td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/1803.07728" target="_blank" rel="noopener noreferrer">Rotation-Pred</a></td><td>selfsup/rotation_pred/r50.py</td><td>default</td><td>r50_lr0_01_head100.py</td><td>18.98</td><td>44.05</td></tr><tr><td><a href="https://arxiv.org/abs/1807.05520" target="_blank" rel="noopener noreferrer">DeepCluster</a></td><td>selfsup/deepcluster/r50.py</td><td>default</td><td>r50_lr0_01_head1_sobel.py</td><td>33.44</td><td>58.62</td></tr><tr><td><a href="https://arxiv.org/abs/1805.01978" target="_blank" rel="noopener noreferrer">NPID</a></td><td>selfsup/npid/r50.py</td><td>default</td><td>r50_lr0_01_head100.py</td><td>27.95</td><td>54.37</td></tr><tr><td><a href="http://openaccess.thecvf.com/content_CVPR_2020/papers/Zhan_Online_Deep_Clustering_for_Unsupervised_Representation_Learning_CVPR_2020_paper.pdf" target="_blank" rel="noopener noreferrer">ODC</a></td><td>selfsup/odc/r50_v1.py</td><td>default</td><td>r50_lr0_1_head100.py</td><td>32.39</td><td>61.02</td></tr><tr><td><a href="https://arxiv.org/abs/1911.05722" target="_blank" rel="noopener noreferrer">MoCo</a></td><td>selfsup/moco/r50_v1.py</td><td>default</td><td>r50_lr0_01_head100.py</td><td>33.15</td><td>61.30</td></tr><tr><td><a href="https://arxiv.org/abs/2003.04297" target="_blank" rel="noopener noreferrer">MoCo v2</a></td><td>selfsup/moco/r50_v2.py</td><td>default</td><td>r50_lr0_01_head100.py</td><td>38.71</td><td>67.90</td></tr><tr><td></td><td>selfsup/moco/r50_v2_simclr_neck.py</td><td>-> SimCLR neck<br></td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/2002.05709" target="_blank" rel="noopener noreferrer">SimCLR</a></td><td>selfsup/simclr/r50_bs256_ep200.py</td><td>default</td><td>r50_lr0_01_head100.py</td><td>36.09</td><td>64.50</td></tr><tr><td></td><td>selfsup/simclr/r50_bs256_ep200_mocov2_neck.py</td><td>-> MoCo v2 neck</td><td></td><td></td><td></td></tr><tr><td><a href="https://arxiv.org/abs/2006.07733" target="_blank" rel="noopener noreferrer">BYOL</a></td><td>selfsup/byol/r50.py</td><td>default</td><td></td><td></td><td></td></tr></tbody></table>
|
||||
|
||||
### PASCAL VOC07+12 Object Detection
|
||||
|
|
|
|||
Loading…
Reference in New Issue