Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/innerlee/gcr
Official repository for ICCV 2023: Get the Best of Both Worlds: Improving Accuracy and Transferability by Grassmann Class
https://github.com/innerlee/gcr
classification geometric-deep-learning grassmann-manifold iccv2023 transfer-learning
Last synced: 5 days ago
JSON representation
Official repository for ICCV 2023: Get the Best of Both Worlds: Improving Accuracy and Transferability by Grassmann Class
- Host: GitHub
- URL: https://github.com/innerlee/gcr
- Owner: innerlee
- Created: 2023-09-08T04:09:02.000Z (about 1 year ago)
- Default Branch: master
- Last Pushed: 2023-10-16T03:10:32.000Z (about 1 year ago)
- Last Synced: 2024-01-27T02:09:21.321Z (9 months ago)
- Topics: classification, geometric-deep-learning, grassmann-manifold, iccv2023, transfer-learning
- Language: Python
- Homepage:
- Size: 2.05 MB
- Stars: 11
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Grassmann Class Representation
[](https://arxiv.org/pdf/2308.01547)
[](./resource/gcr-iccv2023-poster.pdf)
Official code for "Get the Best of Both Worlds: Improving Accuracy and Transferability by Grassmann Class Representation (ICCV 2023)"
https://github.com/innerlee/gcr/assets/9464825/37708eb8-740c-46df-ad24-7bcf6ef592c9
## Installation
Run
```bash
python setup.py develop
```
### Requirements
- pytorch >= 1.13
- mmpretrain
### Prepare data
Put ImageNet1-K dataset on `data/imagenet` folder
## ImageNet1-K Pretrained Weights
| Architecture | Trick | Class Representation | Dim | Top1 | Top5 | Config | Log | Checkpoint |
| ------------ | ------------- | -------------------- | --- | --------- | --------- | ------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------- |
| ResNet50-D | | Vector | - | 78.05 | 93.90 | [config](configs/resnet/resnet50d_8xb32_coslr_in1k.py) | [log](https://drive.google.com/file/d/19NMZs1dlCshW1_8PJx7s200xl_RzoV-j/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1Co5FJYqDcqeEYw_ORTOrRxg3oh2acfRI/view?usp=drive_link) |
| ResNet50-D | | GCR | 1 | 78.42 | 94.14 | [config](configs/resnet/resnet50d_8xb32_coslr_in1k_gcr1.py) | [log](https://drive.google.com/file/d/1e0sJRdtMZOEgM9AfENNsicHhtjMawgw7/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1m1F9TmOQjmKli-_U5ifSxVoiH8R0tHK1/view?usp=drive_link) |
| ResNet50-D | | GCR | 4 | 78.68 | 94.32 | [config](configs/resnet/resnet50d_8xb32_coslr_in1k_gcr4.py) | [log](https://drive.google.com/file/d/12v7UZWrlh3ClIJmz6ysGGMN5TTg50XRI/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1tIXkD9zB7y4Ni3RZYNKgQTZw_PCZkCJV/view?usp=drive_link) |
| ResNet50-D | | GCR | 8 | 79.26 | 94.44 | [config](configs/resnet/resnet50d_8xb32_coslr_in1k_gcr8.py) | [log](https://drive.google.com/file/d/18Hns24rv66163uPKiZZGnoEVal4An7Lc/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1FpuEpzqf-v3c8rsdFX882mKXLY07sXwc/view?usp=drive_link) |
| ResNet50-D | | GCR | 16 | 79.21 | 94.37 | [config](configs/resnet/resnet50d_8xb32_coslr_in1k_gcr16.py) | [log](https://drive.google.com/file/d/1kodUwqZuACdi259DBUVhYlGyMpA-FsVb/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1N9l7Lb-gSMOK-1JzUqa0wLgb81_iBCw1/view?usp=drive_link) |
| ResNet50-D | | GCR | 32 | 78.63 | 94.05 | [config](configs/resnet/resnet50d_8xb32_coslr_in1k_gcr32.py) | [log](https://drive.google.com/file/d/1O0SeintbVELA9_-hIcHIJx930TdlHQ9S/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1Rq2ugIMOZbpDufpiKpewKR4xc9DTExu7/view?usp=drive_link) |
| ResNet50-D | RSB-A1 | Vector | - | 80.53 | 94.98 | [config](configs/resnet/resnet50d_8xb256-rsb-a1-600e_in1k.py) | [log](https://drive.google.com/file/d/1cNw6Kf_BbDn4M3dFfy2gANr3aZgUKEo_/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1ororxl3RMuYTB7TD83nb9Fyq5VuBI5Pq/view?usp=drive_link) |
| ResNet50-D | RSB-A1 | GCR | 8 | 81.00 | 95.40 | [config](configs/resnet/resnet50d_8xb256-rsb-a1-600e_in1k_gcr8.py) | [log](https://drive.google.com/file/d/1SjpT47lTpUPnI33xM9SC3SpC0s440twP/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1njtbpwfz1duOilapCcUnTpGX2eTuTy_U/view?usp=drive_link) |
| ResNet50-D | RSB-A1 FixRes | GCR | 8 | **81.30** | **95.42** | [config](configs/resnet/resnet50d_8xb256-rsb-a1-600e_in1k_gcr8_fixres.py) | [log](https://drive.google.com/file/d/1QSXNiLUK-6VhrktW2hlkNcGS95AJ7RoZ/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/17hYOOMLrg8wOs6g1ehtfbQ_gMWsoYh9j/view?usp=drive_link) |
| ResNet50-D | RSB-A2 | Vector | - | 80.29 | 94.86 | [config](configs/resnet/resnet50d_8xb256-rsb-a2-300e_in1k.py) | [log](https://drive.google.com/file/d/1qaVXUYU7SYhuk5ZeRAblGpVC38XvSv8M/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1LOVRzzP3BfhRvdDgSUgTJ3So9NxllX3N/view?usp=drive_link) |
| ResNet50-D | RSB-A2 | GCR | 8 | 80.75 | 95.24 | [config](configs/resnet/resnet50d_8xb256-rsb-a2-300e_in1k_gcr8.py) | [log](https://drive.google.com/file/d/14BXtMZIM1SEJLdQzRveloE0nRIMcuxH6/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1Vm5xqOXzRXTK_4URXW8rbeevOHXgUOBP/view?usp=drive_link) |
| ResNet50-D | RSB-A2 FixRes | GCR | 8 | 81.04 | 95.46 | [config](configs/resnet/resnet50d_8xb256-rsb-a2-300e_in1k_gcr8_fixres.py) | [log](https://drive.google.com/file/d/17Wyg0r5_a8ENpoPMQOdtXUxUxPoc6P9L/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1RRHAGOd14B7wQb5tcR5blxcm5EBk4wqw/view?usp=drive_link) |
| ResNet50-D | RSB-A3 | Vector | - | 79.36 | 94.47 | [config](configs/resnet/resnet50d_8xb256-rsb-a3-100e_in1k.py) | [log](https://drive.google.com/file/d/1pRxmvOcExofZlhL3ajB99Y7CtQiaC8gD/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1SXSfGEE_OOBDqMYXOFJOmJsj5nTe8Mpy/view?usp=drive_link) |
| ResNet50-D | RSB-A3 | GCR | 8 | 79.64 | 94.85 | [config](configs/resnet/resnet50d_8xb256-rsb-a3-100e_in1k_gcr8.py) | [log](https://drive.google.com/file/d/1x9BGutIOzdvZwvTxSwbu-e7GSWINDXCv/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1gD11fWCReLKU0hoqJzK-7oBZDjWA0S-U/view?usp=drive_link) |
| ResNet50-D | RSB-A3 FixRes | GCR | 8 | 80.20 | 95.15 | [config](configs/resnet/resnet50d_8xb256-rsb-a3-100e_in1k_gcr8_fixres.py) | [log](https://drive.google.com/file/d/1rNwRPSFz_-FLOJxTLnUPSOvSqtrAFBm0/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1ZDXr5V0KIUEk64LrdwTaAvayw7dXD-VL/view?usp=drive_link) |
| ResNet101-D | | Vector | - | 79.31 | 94.67 | [config](configs/resnet/resnet101d_8xb32-coslr_in1k.py) | [log](https://drive.google.com/file/d/1f7si6980G9ni6mC8o2dU5n9yQ-yRtav3/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1RZNdlM00B1TWoOT3YpTD3_TZY0ZDjGdI/view?usp=drive_link) |
| ResNet101-D | | GCR | 8 | 80.24 | 94.95 | [config](configs/resnet/resnet101d_8xb32-coslr_in1k_gcr8.py) | [log](https://drive.google.com/file/d/1PMEOvvh5z602T7HdwfPQCafkVjeXMceJ/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1SS3iyVko63lA6G4OTlDVgVWGmzTV2njF/view?usp=drive_link) |
| ResNet152-D | | Vector | - | 80.00 | 95.02 | [config](configs/resnet/resnet152d_8xb32-coslr_in1k.py) | [log](https://drive.google.com/file/d/1FWxJyDcEpnoRStN16n2GyIevEJ-hHkS1/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1wRIL19r_c8o_2eXZE2ARtEGf77KxhiVn/view?usp=drive_link) |
| ResNet152-D | | GCR | 8 | 80.44 | 95.21 | [config](configs/resnet/resnet152d_8xb32-coslr_in1k_gcr8.py) | [log](https://drive.google.com/file/d/1e4UO1jtTViv7zww6wnR__hinpNSjPx0R/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1yFiJ9lqy9uD1HQ02MWzw_ZlXLiqjK4BY/view?usp=drive_link) |
| VGG13-BN | | Vector | - | 72.02 | 90.79 | [config](configs/vgg/vgg13bn_8xb32_in1k.py) | [log](https://drive.google.com/file/d/1SsLzmDmiQGaDToIYWk-9f8tOp2Mscu3R/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1AP2fI7nFGRZ6mU__E8N-MV5ecEZ1KW4o/view?usp=drive_link) |
| VGG13-BN | | GCR | 8 | 73.40 | 91.30 | [config](configs/vgg/vgg13bn_8xb32_in1k_gcr8.py) | [log](https://drive.google.com/file/d/1tt_iy_R2uSRXm2adPOsLC1uVbij4pEud/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/10EJP9eU-EoFxy11T_qOm8urUHj71BuFz/view?usp=drive_link) |
| Swin-T | | Vector | - | 81.06 | 95.51 | [config](configs/swin_transformer/swin-tiny_8xb128_in1k.py) | [log](https://drive.google.com/file/d/1USGKrheQe6jK1xkFPgUt8M1DFpGc8FNr/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1xstvE_t89F_SNc_zDm2OxRJ-Irq5SLYN/view?usp=drive_link) |
| Swin-T | | GCR | 8 | 81.63 | 95.77 | [config](configs/swin_transformer/swin-tiny_8xb128_in1k_gcr8.py) | [log](https://drive.google.com/file/d/1_mTgAMA95rkwMpzM0hjmM7qNN5V4CNFV/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1RIpMraINw6CjvUAzNiAJOm11xQq-SUSx/view?usp=drive_link) |
| Deit3-S | | Vector | - | 81.53 | 95.22 | [config](configs/deit3/deit3-small-p16_8xb256_in1k-224px.py) | [log](https://drive.google.com/file/d/11FEoYMfzTxOXeFAubz-BsnEhv7mCrXYW/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1WFPbxiOOu380qsYQeNFtRSwVJV4eZR12/view?usp=drive_link) |
| Deit3-S | | GCR | 8 | 82.18 | 95.73 | [config](configs/deit3/deit3-small-p16_8xb256_in1k-224px_gcr8.py) | [log](https://drive.google.com/file/d/1r-mJZ04DFcJwMfDij9fSR7eEh16cyZeK/view?usp=drive_link) | [ckpt](https://drive.google.com/file/d/1n-uHEWuiyJ1OLeuD555hfyJugKmXBYCx/view?usp=drive_link) |
## Training
```bash
# train ResNet50d baseline with 8 gpus
./tools/dist_train.sh configs/resnet/resnet50d_8xb32-coslr_in1k.py 8
# train the gcr version
./tools/dist_train.sh configs/resnet/resnet50d_8xb32-coslr_in1k_gcr8.py 8
```
## Citation
```
@inproceedings{haoqi2023gcr,
title = {Get the Best of Both Worlds: Improving Accuracy and Transferability by Grassmann Class Representation},
author = {Wang, Haoqi and Li, Zhizhong and Zhang, Wayne},
booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
year = {2023}
}
```
## Related Projects
[ViM: Out-Of-Distribution with Virtual-logit Matching (CVPR 2022)](https://github.com/haoqiwang/vim)