# Seesaw Loss for Long-Tailed Instance Segmentation (CVPR 2021) ## Introduction We provide config files to reproduce the instance segmentation performance in the CVPR 2021 paper for [Seesaw Loss for Long-Tailed Instance Segmentation](https://arxiv.org/abs/2008.10032). ```latex @inproceedings{wang2021seesaw, title={Seesaw Loss for Long-Tailed Instance Segmentation}, author={Jiaqi Wang and Wenwei Zhang and Yuhang Zang and Yuhang Cao and Jiangmiao Pang and Tao Gong and Kai Chen and Ziwei Liu and Chen Change Loy and Dahua Lin}, booktitle={Proceedings of the {IEEE} Conference on Computer Vision and Pattern Recognition}, year={2021} } ``` * Please setup [LVIS dataset](../lvis/README.md) for MMDetection. * RFS indicates to use oversample strategy [here](../../docs/tutorials/customize_dataset.md#class-balanced-dataset) with oversample threshold `1e-3`. ## Results and models of Seasaw Loss on LVIS v1 dataset | Method | Backbone | Style | Lr schd | Data Sampler | Norm Mask | box AP | mask AP | Config | Download | | :----------------: | :-------: | :-----: | :-----: | :----------: | :-------: | :----: | :-----: | :---------------------------------------------------------------------------------------------------------------------------------------------------------------------: | :------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: | | Mask R-CNN | R-50-FPN | pytorch | 2x | random | N | 25.6 | 25.0 | [config](https://github.com/open-mmlab/mmdetection/tree/master/configs/seesaw_loss/mask_rcnn_r50_fpn_random_seesaw_loss_mstrain_2x_lvis_v1.py) | [model](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/mask_rcnn_r50_fpn_random_seesaw_loss_mstrain_2x_lvis_v1-a698dd3d.pth) | [log](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/mask_rcnn_r50_fpn_random_seesaw_loss_mstrain_2x_lvis_v1.log.json) | | Mask R-CNN | R-50-FPN | pytorch | 2x | random | Y | 25.6 | 25.4 | [config](https://github.com/open-mmlab/mmdetection/tree/master/configs/seesaw_loss/mask_rcnn_r50_fpn_random_seesaw_loss_normed_mask_mstrain_2x_lvis_v1.py) | [model](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/mask_rcnn_r50_fpn_random_seesaw_loss_normed_mask_mstrain_2x_lvis_v1-a1c11314.pth) | [log](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/mask_rcnn_r50_fpn_random_seesaw_loss_normed_mask_mstrain_2x_lvis_v1.log.json) | | Mask R-CNN | R-101-FPN | pytorch | 2x | random | N | 27.4 | 26.7 | [config](https://github.com/open-mmlab/mmdetection/tree/master/configs/mask_rcnn_r101_fpn_random_seesaw_loss_mstrain_2x_lvis_v1.py) | [model](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/mask_rcnn_r101_fpn_random_seesaw_loss_mstrain_2x_lvis_v1-8e6e6dd5.pth) | [log](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/mask_rcnn_r101_fpn_random_seesaw_loss_mstrain_2x_lvis_v1.log.json) | | Mask R-CNN | R-101-FPN | pytorch | 2x | random | Y | 27.2 | 27.3 | [config](https://github.com/open-mmlab/mmdetection/tree/master/configs/seesaw_loss/mask_rcnn_r101_fpn_random_seesaw_loss_normed_mask_mstrain_2x_lvis_v1.py) | [model](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/mask_rcnn_r101_fpn_random_seesaw_loss_normed_mask_mstrain_2x_lvis_v1-a0b59c42.pth) | [log](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/mask_rcnn_r101_fpn_random_seesaw_loss_normed_mask_mstrain_2x_lvis_v1.log.json) | | Mask R-CNN | R-50-FPN | pytorch | 2x | RFS | N | 27.6 | 26.4 | [config](https://github.com/open-mmlab/mmdetection/tree/master/configs/seesaw_loss/mask_rcnn_r50_fpn_sample1e-3_seesaw_loss_mstrain_2x_lvis_v1.py) | [model](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/mask_rcnn_r50_fpn_sample1e-3_seesaw_loss_mstrain_2x_lvis_v1-392a804b.pth) | [log](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/mask_rcnn_r50_fpn_sample1e-3_seesaw_loss_mstrain_2x_lvis_v1.log.json) | | Mask R-CNN | R-50-FPN | pytorch | 2x | RFS | Y | 27.6 | 26.8 | [config](https://github.com/open-mmlab/mmdetection/tree/master/configs/seesaw_loss/mask_rcnn_r50_fpn_sample1e-3_seesaw_loss_normed_mask_mstrain_2x_lvis_v1.py) | [model](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/mask_rcnn_r50_fpn_sample1e-3_seesaw_loss_normed_mask_mstrain_2x_lvis_v1-cd0f6a12.pth) | [log](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/mask_rcnn_r50_fpn_sample1e-3_seesaw_loss_normed_mask_mstrain_2x_lvis_v1.log.json) | | Mask R-CNN | R-101-FPN | pytorch | 2x | RFS | N | 28.9 | 27.6 | [config](https://github.com/open-mmlab/mmdetection/tree/master/configs/seesaw_loss/mask_rcnn_r101_fpn_sample1e-3_seesaw_loss_mstrain_2x_lvis_v1.py) | [model](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/mask_rcnn_r101_fpn_sample1e-3_seesaw_loss_mstrain_2x_lvis_v1-e68eb464.pth) | [log](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/mask_rcnn_r101_fpn_sample1e-3_seesaw_loss_mstrain_2x_lvis_v1.log.json) | | Mask R-CNN | R-101-FPN | pytorch | 2x | RFS | Y | 28.9 | 28.2 | [config](https://github.com/open-mmlab/mmdetection/tree/master/configs/seesaw_loss/mask_rcnn_r101_fpn_sample1e-3_seesaw_loss_normed_mask_mstrain_2x_lvis_v1.py) | [model](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/mask_rcnn_r101_fpn_sample1e-3_seesaw_loss_normed_mask_mstrain_2x_lvis_v1-1d817139.pth) | [log](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/mask_rcnn_r101_fpn_sample1e-3_seesaw_loss_normed_mask_mstrain_2x_lvis_v1.log.json) | | Cascade Mask R-CNN | R-101-FPN | pytorch | 2x | random | N | 33.1 | 29.2 | [config](https://github.com/open-mmlab/mmdetection/tree/master/configs/seesaw_loss/cascade_mask_rcnn_r101_fpn_random_seesaw_loss_mstrain_2x_lvis_v1.py) | [model](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/cascade_mask_rcnn_r101_fpn_random_seesaw_loss_mstrain_2x_lvis_v1-71e2215e.pth) | [log](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/cascade_mask_rcnn_r101_fpn_random_seesaw_loss_mstrain_2x_lvis_v1.log.json) | | Cascade Mask R-CNN | R-101-FPN | pytorch | 2x | random | Y | 33.0 | 30.0 | [config](https://github.com/open-mmlab/mmdetection/tree/master/configs/seesaw_loss/cascade_mask_rcnn_r101_fpn_random_seesaw_loss_normed_mask_mstrain_2x_lvis_v1.py) | [model](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/cascade_mask_rcnn_r101_fpn_random_seesaw_loss_normed_mask_mstrain_2x_lvis_v1-8b5a6745.pth) | [log](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/cascade_mask_rcnn_r101_fpn_random_seesaw_loss_normed_mask_mstrain_2x_lvis_v1.log.json) | | Cascade Mask R-CNN | R-101-FPN | pytorch | 2x | RFS | N | 30.0 | 29.3 | [config](https://github.com/open-mmlab/mmdetection/tree/master/configs/seesaw_loss/cascade_mask_rcnn_r101_fpn_sample1e-3_seesaw_loss_mstrain_2x_lvis_v1.py) | [model](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/cascade_mask_rcnn_r101_fpn_sample1e-3_seesaw_loss_mstrain_2x_lvis_v1-5d8ca2a4.pth) | [log](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/cascade_mask_rcnn_r101_fpn_sample1e-3_seesaw_loss_mstrain_2x_lvis_v1.log.json) | | Cascade Mask R-CNN | R-101-FPN | pytorch | 2x | RFS | Y | 32.8 | 30.1 | [config](https://github.com/open-mmlab/mmdetection/tree/master/configs/seesaw_loss/cascade_mask_rcnn_r101_fpn_sample1e-3_seesaw_loss_normed_mask_mstrain_2x_lvis_v1.py) | [model](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/cascade_mask_rcnn_r101_fpn_sample1e-3_seesaw_loss_normed_mask_mstrain_2x_lvis_v1-c8551505.pth) | [log](https://download.openmmlab.com/mmdetection/v2.0/seesaw_loss/cascade_mask_rcnn_r101_fpn_sample1e-3_seesaw_loss_normed_mask_mstrain_2x_lvis_v1.log.json) |