ref:GitHub - SJTU-Thinklab-Det/DOTA-DOAI: This repo is the codebase for our team to participate in DOTA related competitions, including rotation and horizontal detection.
Abstract
This repo is the codebase for our team to participate in DOTA related competitions, including rotation and horizontal detection. We mainly use FPN-based two-stage detector, and it is completed by YangXue and YangJirui.
We also recommend a tensorflow-based rotation detection benchmark, which is led by YangXue.
Performance
DOTA1.0 (Task1)
Model | Backbone | Training data | Val data | mAP | Model Link | Tricks | lr schd | Data Augmentation | GPU | Image/GPU | Configs |
FPN | ResNet152_v1d (600,800,1024)->MS | DOTA1.0 trainval | DOTA1.0 test | 78.99 | model | ALL | 2x | Yes | 2X GeForce RTX 2080 Ti | 1 | cfgs_dota1.0_res152_v1.py |
### DOTA1.0 (Task2) | |||||||||||
Model | Backbone | Training data | Val data | mAP | Model Link | Tricks | lr schd | Data Augmentation | GPU | Image/GPU | Configs |
:------------: | :------------: | :---------: | :-----------: | :----------: | :-----------: | :---------: | :---------: | :---------: | :---------: | :---------: | :---------: |
FPN (memory consumption) | ResNet152_v1d (600,800,1024)->MS | DOTA1.0 trainval | DOTA1.0 test | 81.23 | model | ALL | 2x | Yes | 2X Quadro RTX 8000 | 1 | cfgs_dota1.0_res152_v1.py |
Visualization
Performance of published papers on DOTA datasets
DOTA1.0 (Task1)
Model | Backbone | mAP | Paper Link | Code Link | Remark | Recommend |
FR-O (DOTA) | ResNet101 | 52.93 | CVPR2018 | MXNet | DOTA dataset, baseline | ✅ |
IENet | ResNet101 | 57.14 | arXiv:1912.00969 | - | anchor free | |
TOSO | ResNet101 | 57.52 | ICASSP2020 | - | geometric transformation | |
PIoU Loss | DLA-34 | 60.5 | ECCV2020 | Pytorch | IoU loss, anchor free | ✅ |
R2CNN | ResNet101 | 60.67 | arXiv:1706.09579 | TF | scene text, multi-task, different pooled sizes, baseline | ✅ |
RRPN | ResNet101 | 61.01 | TMM arXiv:1703.01086 | TF | scene text, rotation proposals, baseline | ✅ |
Axis Learning | ResNet101 | 65.98 | Remote Sensing | - | single stage, anchor free | ✅ |
MARNet | ResNet101 | 67.19 | IJRS | - | based on scrdet | |
ICN | ResNet101 | 68.16 | ACCV2018 | - | image cascade, multi-scale | ✅ |
GSDet | ResNet101 | 68.28 | TIP | - | scale reasoning | |
RADet | ResNeXt101 | 69.09 | Remote Sensing | - | enhanced FPN, mask rcnn | |
KARNET | ResNet50 | 68.87 | CISNRC 2020 | - | attention denoising, anchor refining | |
RoI Transformer | ResNet101 | 69.56 | CVPR2019 | MXNet, Pytorch | roi transformer | ✅ |
CAD-Net | ResNet101 | 69.90 | TGRS arXiv:1903.00857 | - | attention | |
ProbIoU | ResNet50 | 70.04 | arXiv:2106.06072 | TF | gaussian bounding boxes, hellinger distance | ✅ |
A2S-Det | ResNet101 | 70.64 | Remote Sensing | - | label assign | |
AOOD | ResNet101 | 71.18 | Neural Computing and Applications | - | attention + R-DFPN | |
Cascade-FF | ResNet152 | 71.80 | ICME2020 | - | refined retinanet + feature fusion | |
SCPNet | Hourglass104 | 72,20 | GRSL | - | corner points | |
P-RSDet | ResNet101 | 72.30 | Access | - | anchor free, polar coordinates | ✅ |
BBAVectors | ResNet101 | 72.32 | WACV2021 | Pytorch | keypoint based | ✅ |
ROPDet | ResNet101-DCN | 72.42 | J REAL-TIME IMAGE PR | - | point set representation | |
SCRDet | ResNet101 | 72.61 | ICCV2019 | TF: R2CNN++, IoU-Smooth L1: RetinaNet-based, R3Det-based | attention, angular boundary problem | ✅ |
O2-DNet | Hourglass104 | 72.8 | ISPRS, arXiv:1912.10694 | - | centernet, anchor free | ✅ |
HRPNet | HRNet-W48 | 72.83 | GRSL | - | polar | |
SARD | ResNet101 | 72.95 | Access | - | IoU-based weighted loss | |
GLS-Net | ResNet101 | 72.96 | Remote Sensing | - | attention, saliency pyramid | |
ProjBB | ResNet101 | 73.03 | Access | code, codebase | new definition of bounding box | |
DRN | Hourglass104 | 73.23 | CVPR(oral) | code | centernet, feature selection module, dynamic refinement head, new dataset (SKU110K-R) | ✅ |
FADet | ResNet101 | 73.28 | ICIP2019 | - | attention | |
MFIAR-Net | ResNet152 | 73.49 | Sensors | - | feature attention, enhanced FPN | |
CFC-NET | ResNet101 | 73.50 | arXiv:2101.06849 | Pytorch | critical feature, label assign, refine | ✅ |
R3Det | ResNet152 | 73.74 | AAAI2021 | TF, Pytorch | refined single stage, feature alignment | ✅ |
RSDet | ResNet152 | 74.10 | AAAI2021 | TF | quadrilateral bbox, angular boundary problem | ✅ |
SegmRDet | ResNet50 | 74.14 | Neurocomputing | - | segmentation-baed, new training and inference mechanism | |
Gliding Vertex | ResNet101 | 75.02 | TPAMI arXiv:1911.09358 | Pytorch | quadrilateral bbox | ✅ |
EFN | U-Net | 75.27 | Preprints | - | Field-based | ✅ |
SAR | ResNet152 | 75.26 | Access | - | boundary problem | ✅ |
TricubeNet | Hourglass104 | 75.26 | arXiv:2104.11435 | code | 2D tricube kernel | ✅ |
Mask OBB | ResNeXt-101 | 75.33 | Remote Sensing | - | attention, multi-task | ✅ |
- | DarkNet | 75.5 | TGRS | - | angle classification | |
FFA | ResNet101 | 75.7 | ISPRS | - | enhanced FPN, rotation proposals | |
CBDA-Net | DLA-34-DCN | 75.74 | TGRS | - | dual attention | |
APE | ResNeXt-101(32x4) | 75.75 | TGRS arXiv:1906.09447 | - | adaptive period embedding, length independent IoU (LIIoU) | ✅ |
R4Det | ResNet152 | 75.54 | Image Vis Comput | - | feature recursion and refinement | |
F3-Net | ResNet152 | 76.02 | Remote Sensing | - | feature fusion and filtration | |
CenterMap OBB | ResNet101 | 76.03 | TGRS | - | center-probability-map | |
CSL | ResNet152 | 76.17 | ECCV2020 | TF: CSL_RetinaNet, Pytorch: YOLOv5_DOTA_OBB (CSL) | angular boundary problem | ✅ |
MRDet | ResNet101 | 76.24 | arXiv:2012.13135 | - | arbitrary-oriented rpn, multiple subtasks | |
AFC-Net | ResNet101 | 76.27 | Neurocomputing | - | adaptive feature concatenate | |
OWSR | Ensemble (ResNet101 + ResNeXt101 + mdcn-ResNet101) | 76.36 | CVPR2019 WorkShop TGRS | - | enhanced FPN | |
OPLD | ResNet101 | 76.43 | J-STARS | Pytorch | boundary problem, point-guided | ✅ |
R3Det++ | ResNet152 | 76.56 | arXiv:2004.13316 | TF | refined single stage, feature alignment, denoising | ✅ |
PolarDet | ResNet101 | 76.64 | IJRS arXiv:2010.08720 | - | polar, center-semantic | ✅ |
Beyond Bounding-Box | ResNet152 | 76.67 | CVPR2021 | Pytorch | point-based, reppoints | ✅ |
SCRDet++ | ResNet101 | 76.81 | arXiv:2004.13316 | TF | angular boundary problem, denoising | ✅ |
DAL+S2A-Net | ResNet50 | 76.95 | AAAI2021 | Pytorch | label assign | ✅ |
DCL | ResNet152 | 77.37 | CVPR2021 | TF | boundary problem | ✅ |
MSFF | ResNet50 | 77.46 | JSTARS | - | rotation invariance features | |
RIDet | ResNet50 | 77.62 | arXiv:2103.11636 | Pytorch, TF | quad., representation ambiguity | ✅ |
RDD | ResNet101 | 77.75 | Remote Sensing | Pytorch | rotation-decoupled | |
OSKDet | ResNet101 | 77.81 | arXiv:2104.08697 | - | keypoint localization (very similar to FR-Est) | |
CG-Net | ResNet101 | 77.89 | arXiv:2103.11399 | Pytorch | attention | |
Oriented RepPoints | ResNet101 | 78.12 | arXiv:2105.11111 | Pytorch | point-based, reppoints | ✅ |
FR-Est | ResNet101-DCN | 78.49 | TGRS | - | point-based estimator | ✅ |
S2A-Net | ResNet50/ResNet101 | 79.42/79.15 | TGRS | Pytorch | refined single stage, feature alignment | ✅ |
O2DETR | ResNet50 | 79.66 | arXiv:2106.03146 | - | deformable detr, transformer | ✅ |
ROSD | ResNet101 | 79.76 | Access | - | refined single stage, feature alignment | |
SARA | ResNet50/ResNet101 | 79.91/79.13 | Remote Sensing | - | self-adaptive aspect ratio anchor, refine | |
ReDet | ReR50-ReFPN | 80.10 | CVPR2021 | Pytorch | rotation-equivariant, rotation-invariant roI align | ✅ |
GWD | ResNet152 | 80.23 | ICML2021 | TF | boundary discontinuity, square-like problem, gaussian wasserstein distance loss | ✅ |
KLD | ResNet152 | 80.63 | arXiv:2106.01883 | TF | Kullback-Leibler divergence, high-precision, scale invariance | ✅ |
DOTA1.0 (Task2)
Model | Backbone | mAP | Paper Link | Code Link | Remark | Recommend |
FR-H (DOTA) | ResNet101 | 60.46 | CVPR2018 | MXNet | DOTA dataset, baseline | ✅ |
Deep Active Learning | ResNet18 | 64.26 | arXiv:2003.08793 | - | CenterNet, Deep Active Learning | ✅ |
SBL | ResNet50 | 64.77 | arXiv:1810.08103 | - | single stage | |
CenterFPANet | ResNet18 | 65.29 | HPCCT & BDAI 2020 arXiv:2009.03063 | - | light-weight | |
MARNet | ResNet101 | 71.73 | IJRS | - | based on scrdet | |
FMSSD | VGG16 | 72.43 | TGRS | - | IoU-based weighted loss, enhanced FPN | |
ICN | ResNet101 | 72.45 | ACCV2018 | - | image cascade, multi-scale | ✅ |
IoU-Adaptive R-CNN | ResNet101 | 72.72 | Remote Sensing | - | IoU-based weighted loss, cascade | |
EFR | VGG16 | 73.49 | Remote Sensing | Pytorch | enhanced FPN | |
AF-EMS | ResNet101 | 73.97 | Remote Sensing | - | scale-aware feature, anchor free | |
SCRDet | ResNet101 | 75.35 | ICCV2019 | TF | attention, angular boundary problem | ✅ |
FADet | ResNet101 | 75.38 | ICIP2019 | - | attention | |
MFIAR-Net | ResNet152 | 76.07 | Sensors | - | feature attention, enhanced FPN | |
F3-Net | ResNet152 | 76.48 | Remote Sensing | - | feature fusion and filtration | |
Mask OBB | ResNeXt-101 | 76.98 | Remote Sensing | - | attention, multi-task | ✅ |
CenterMap OBB | ResNet101 | 77.33 | TGRS | - | center-probability-map | |
ASSD | VGG16 | 77.8 | TGRS | - | feature aligned | |
AFC-Net | ResNet101 | 78.06 | Neurocomputing | - | adaptive feature concatenate | |
CG-Net | ResNet101 | 78.26 | arXiv:2103.11399 | Pytorch | attention | |
OPLD | ResNet101 | 78.35 | J-STARS | Pytorch | boundary problem, point-guided | ✅ |
A2RMNet | ResNet101 | 78.45 | Remote Sensing | - | attention, enhanced FPN, different pooled sizes | |
OWSR | Ensemble (ResNet101 + ResNeXt101 + mdcn-ResNet101) | 78.79 | CVPR2019 WorkShop TGRS | - | enhanced FPN | |
Parallel Cascade R-CNN | ResNeXt-101 | 78.96 | Journal of Physics: Conference Series | - | cascade rcnn | |
DM-FPN | ResNet-Based | 79.27 | Remote Sensing | - | enhanced FPN | |
SCRDet++ | ResNet101 | 79.35 | arXiv:2004.13316 | TF | denoising | ✅ |
DOTA1.5 (Task1)
Model | Backbone | mAP | Paper Link | Code Link | Remark | Recommend |
APE | ResNeXt-101(32x4) | 78.34 | TGRS arXiv:1906.09447 | - | length independent IoU (LIIoU) | ✅ |
OWSR | Ensemble (ResNet101 + ResNeXt101 + mdcn-ResNet101) | 76.60 | TGRS CVPR2019 WorkShop | - | enhanced FPN | |
ReDet | ReR50-ReFPN | 76.80 | CVPR2021 | Pytorch | rotation-equivariant, rotation-invariant RoI Align, | ✅ |
DOTA1.5 (Task2)
Model | Backbone | mAP | Paper Link | Code Link | Remark | Recommend |
CDD-Net | ResNet101 | 61.3 | GRSL | - | attention | |
ReDet | ReR50-ReFPN | 78.08 | CVPR2021 | Pytorch | rotation-equivariant, rotation-invariant RoI Align, | ✅ |
OWSR | Ensemble (ResNet101 + ResNeXt101 + mdcn-ResNet101) | 79.50 | TGRS CVPR2019 WorkShop | - | enhanced FPN |
Related Articles
Model | Paper Link | Code Link | Remark | Recommend |
SSSDET | ICIP2019 arXiv:1909.00292 | - | vehicle detection, lightweight | |
AVDNet | GRSL arXiv:1907.07477 | - | vehicle detection, small object | |
ClusDet | ICCV2019 | Caffe2 | object cluster regions | ✅ |
DMNet | CVPR2020 WorkShop | - | object cluster regions | ✅ |
AdaZoom | arXiv:2106.10409 | - | object cluster regions, reinforcement learning | ✅ |
OIS | arXiv:1911.07732 | related Pytorch code | Oriented Instance Segmentation | ✅ |
ISOP | IGARSS2020 | - | Oriented Instance Segmentation | |
LR-RCNN | arXiv:2005.14264 | - | vehicle detection | - |
GRS-Det | TGRS | - | ship detection, rotation fcos | - |
DRBox | arXiv:1711.09405 | Caffe | sar object detection | ✅ |
DRBox-v2 | TGRS | TF | sar object detection | - |
RAPiD | arXiv:2005.11623 | Pytorch | overhead fisheye images | - |
OcSaFPN | arXiv:2012.09859 | - | denoising | - |
CR2A-Net | TGRS | - | ship detection | - |
- | TGRS | - | knowledge distillation | ✅ |
CHPDet | arXiv:2101.11189 | - | new ship dataset | ✅ |
Other Rotation Detection Codes
Base Method | Code Link |
RetinaNet | RetinaNet_Tensorflow_Rotation |
YOLOv3 | rotate-yolov3-Pytorch, YOLOv3-quadrangle-Pytorch, yolov3-polygon-Pytorch |
YOLOv5 | rotation-yolov5-Pytorch, YOLOv5_DOTA_OBB (CSL) |
CenterNet | R-CenterNet-Pytorch |
Dataset
Name | Categories | Annotation | Paper | Download | Remark |
DOTA1.0 | 15 | oriented BB | CVPR2018 | Link | |
DOTA1.5 | 16 | oriented BB | CVPR2018 | Link | |
DOTA2.0 | 18 | oriented BB | CVPR2018 | Link | |
iSAID | 15 | instance | CVPRW2019 | Link | |
AI-TOD | 8 | horizontal BB | ICPR2021 | Link | |
DIOR | 20 | horizontal BB | ISPRS | Baidu Drive (ibhm) | |
NWPU VHR-10 | 10 | horizontal BB | TGRS | Link | |
UCAS-AOD | 2 | oriented BB | ICIP | Link, Baidu Drive (r2mr) | |
UAV-ROD | 1 | oriented BB | - | Link | Car |
HRRSD | 13 | horizontal BB | TGRS | Link | |
RSOD | 4 | horizontal BB | TGRS | Link | |
SAR-Ship-Dataset | 1 | horizontal BB | Remote Sensing | Link | SAR Ship |
SSDD | 1 | horizontal BB | BIGSARDATA | Baidu Drive (fyh0) | SAR Ship |
SSDD+ | 1 | oriented BB | - | Baidu Drive (oh6x) | SAR Ship |
AIR-SARShip-1.0 | 1 | horizontal BB | 雷达学报 | Link | SAR Ship |
HRSID | 1 | instance | - | Link | SAR Ship |
HRSC2016 | 4 | oriented BB | ICPR | Baidu Drive (rfg6) | Ship |
FGSD | 4 | oriented BB | arXiv:2003.06832 | - | Ship |
FGSD2021 | 20 | oriented BB | arXiv:2101.11189 | - | Ship |
DLR-3K | 2 | oriented BB | GRSL | Baidu Drive (bh71) | Vehicle |
VEDAI | 9 | oriented BB | JVCIR | Link | Vehicle |
COWC | 1 | one dot | ECCV2016 | Link | Vehicle |
UVSD | 1 | instance | Remote Sensing | Link | Vehicle |
EAGLE | 2 | oriented BB | arXiv:2007.06124 | Link | Vehicle |
RarePlanes | 1 to 110 | instance | arXiv:2006.02963 | Link | Plane |
For more remote sensing datasets of different research directions, please visit here.