This repository is the code that needs to be submitted for OpenMMLab Algorithm Ecological Challenge, the paper is BoxInst: High-Performance Instance Segmentation with Box Annotations
This project is released under the Apache 2.0 license.
Name | box AP | mask AP | log | download |
---|---|---|---|---|
condInst_MS_R_50_1x | 0.393 | 0.349 | log | model |
Name | box AP | mask AP | log | download |
---|---|---|---|---|
BoxInst_MS_R_50_1x | 0.390 | 0.304 | log | model |
BoxInst_MS_R_50_2x | 0.408 | 0.313 | log | model |
BoxInst_MS_R_50_90k | 0.388 | 0.302 | log | model |
BoxInst_MS_R_101_1x | 0.412 | 0.318 | log | model |
BoxInst_MS_R_101_90k | 0.410 | 0.318 | - | model |
Some other methods in MMDetection are also supported.
Our project is totally based on MMCV and MMDetection. Please see get_started.md for the basic usage of MMDetection.
Please see doc to start training. Example,
CUDA_VISIBLE_DEVICES=0,1,2,3 PORT=29500 sh ./tools/dist_train.sh configs/boxinst/boxinst_r50_caffe_fpn_coco_mstrain_1x.py 4
please following linear linear scaling rule to adjust batch size, learning rate and iterations.
python tools/test.py configs/boxinst/boxinst_r50_caffe_fpn_coco_mstrain_1x.py work_dirs/boxinst_r50_caffe_fpn_coco_mstrain_1x.py/latest.pth --eval bbox segm
boxinst_r50_caffe_fpn_coco_mstrain_2x is added! Longer training better performance!
OBBDetection is an oriented object detection toolbox based on MMdetection. What an awesome work from jbwang1997!!!
- MMCV: OpenMMLab foundational library for computer vision.
- MMDetection: OpenMMLab detection toolbox and benchmark.
- Adelaidet