DenseSIRST
收藏资源简介:
DenseSIRST是由南京理工大学计算机科学与工程学院创建的红外小目标检测数据集,专注于密集目标检测。该数据集提供了背景区域的像素级语义标注,支持从稀疏到密集目标检测的转变。数据集内容包括密集的小目标及其背景的详细分割,旨在解决复杂背景下的目标检测问题,特别是减少误报率。通过精细的背景语义模型,DenseSIRST支持开发更有效的检测算法。
DenseSIRST is an infrared small target detection dataset created by the School of Computer Science and Engineering, Nanjing University of Science and Technology, which focuses on dense target detection. This dataset provides pixel-level semantic annotations for background regions, enabling the transition from sparse to dense target detection. The dataset includes detailed segmentations of dense small targets and their backgrounds, aiming to address the problem of target detection under complex backgrounds, particularly reducing the false positive rate. Through fine-grained background semantic models, DenseSIRST supports the development of more effective detection algorithms.
数据集概述
数据集名称
DenseSIRST Datasets
数据集链接
文件结构
angular2html |- data |- SIRSTdevkit |- PNGImages |- Misc_1.png ...... |- SIRST |- BBox |- Misc_1.xml ...... |- BinaryMask |- Misc_1_pixels0.png |- Misc_1.xml ...... |- PaletteMask |- Misc_1.png ...... |- Point_label |- Misc_1_pixels0.txt ...... |- SkySeg |- BinaryMask |- Misc_1_pixels0.png |- Misc_1.xml ...... |- PaletteMask |- Misc_1.png ...... |- Splits |- train_v2.txt |- test_v2.txt ......
PNGImages:存储所有图像的文件夹。SIRST和SkySeg:存储标注文件的文件夹。SIRST对应红外小目标。SkySeg对应天空分割。
训练和测试
训练命令
shell $ CUDA_VISIBLE_DEVICES=0 python train.py <CONFIG_FILE>
示例: shell $ CUDA_VISIBLE_DEVICES=0 python tools/train_det.py configs/detection/fcos_changer_seg/fcos_changer_seg_r50-caffe_fpn_gn-head_1x_densesirst.py
测试命令
shell $ CUDA_VISIBLE_DEVICES=0 python test.py <CONFIG_FILE> <SEG_CHECKPOINT_FILE>
示例: shell $ CUDA_VISIBLE_DEVICES=0 python tools/test_det.py configs/detection/fcos_changer_seg/fcos_changer_seg_r50-caffe_fpn_gn-head_1x_densesirst.py work_dirs/fcos_changer_seg_r50-caffe_fpn_gn-head_1x_densesirst/20240719_162542/best_pascal_voc_mAP_epoch_8.pth
若要可视化结果,可在命令末尾添加 --show。
模型库和基准测试
排行榜
| 方法 | 主干网络 | mAP<sub>07</sub>↑ | recall<sub>07</sub>↑ | mAP<sub>12</sub>↑ | recall<sub>12</sub>↑ | Flops↓ | Params↓ |
|---|---|---|---|---|---|---|---|
| One-stage | |||||||
| FCOS | ResNet50 | 0.232 | 0.315 | 0.204 | 0.324 | 50.291G | 32.113M |
| SSD | 0.211 | 0.421 | 0.178 | 0.424 | 87.552G | 23.746M | |
| GFL | ResNet50 | 0.253 | 0.332 | 0.230 | 0.317 | 52.296G | 32.258M |
| ATSS | ResNet50 | 0.248 | 0.327 | 0.202 | 0.326 | 51.504G | 32.113M |
| CenterNet | ResNet50 | 0.000 | 0.000 | 0.000 | 0.000 | 50.278G | 32.111M |
| PAA | ResNet50 | 0.255 | 0.545 | 0.228 | 0.551 | 51.504G | 32.113M |
| PVT-T | 0.109 | 0.481 | 0.093 | 0.501 | 41.623G | 21.325M | |
| RetinaNet | ResNet50 | 0.114 | 0.510 | 0.086 | 0.523 | 52.203G | 36.330M |
| EfficientDet | 0.099 | 0.433 | 0.072 | 0.419 | 34.686G | 18.320M | |
| TOOD | ResNet50 | 0.256 | 0.355 | 0.226 | 0.342 | 50.456G | 32.018M |
| VFNet | ResNet50 | 0.253 | 0.336 | 0.214 | 0.336 | 48.317G | 32.709M |
| YOLOF | ResNet50 | 0.091 | 0.009 | 0.002 | 0.009 | 25.076G | 42.339M |
| AutoAssign | ResNet50 | 0.255 | 0.354 | 0.180 | 0.314 | 50.555G | 36.244M |
| DyHead | ResNet50 | 0.249 | 0.335 | 0.189 | 0.328 | 27.866G | 38.890M |
| Two-stage | |||||||
| Faster R-CNN | ResNet50 | 0.091 | 0.022 | 0.015 | 0.029 | 0.759T | 33.035M |
| Cascade R-CNN | ResNet50 | 0.136 | 0.188 | 0.139 | 0.194 | 90.978G | 69.152M |
| Dynamic R-CNN | ResNet50 | 0.184 | 0.235 | 0.111 | 0.190 | 63.179G | 41.348M |
| Grid R-CNN | ResNet50 | 0.091 | 0.018 | 0.025 | 0.037 | 0.177T | 64.467M |
| Libra R-CNN | ResNet50 | 0.141 | 0.142 | 0.085 | 0.120 | 63.990G | 41.611M |
| End2End | |||||||
| DETR | ResNet50 | 0.000 | 0.000 | 0.000 | 0.000 | 24.940G | 41.555M |
| Deformable DETR | ResNet50 | 0.024 | 0.016 | 0.018 | 0.197 | 51.772G | 40.099M |
| DAB-DETR | ResNet50 | 0.005 | 0.054 | 0.000 | 0.001 | 28.939G | 43.702M |
| Conditional DETR | ResNet50 | 0.000 | 0.000 | 0.000 | 0.001 | 27.143G | 40.297M |
| Sparse R-CNN | ResNet50 | 0.183 | 0.572 | 0.154 | 0.614 | 45.274G | 0.106G |
| BAFE-Net (Ours) | ResNet50 | 0.270 | 0.332 | 0.236 | 0.329 | 69.114G | 35.329M |
模型库

- 1Background Semantics Matter: Cross-Task Feature Exchange Network for Clustered Infrared Small Target Detection With Sky-Annotated Dataset南京理工大学计算机科学与工程学院 · 2024年



