Chinese Journal of Liquid Crystals and Displays, Volume. 40, Issue 3, 516(2025)
Few-shot wildlife detection based on multi-scale context extraction
Fig. 6. Examples of FSWA dataset. （a） Leopard；（b） Ostrich； （c） Panda； （d） Rhinoceros； （e） Kangaroo.
Fig. 8. Comparison of some test results. （a） TFA； （b） FSDetView； （c） FSCE； （d） DeFRCN； （e） VFA； （f） MS-FSWD.
Table 1. FSWA dataset statistics
View tableView in ArticleTable 1. FSWA dataset statistics
动物种类 图像数量 实例数量 Bird 813 1 288 Cat 1 127 1 276 Cow 340 771 Dog 1 341 1 598 Horse 526 803 Sheep 357 1 084 Leopard 260 298 Ostrich 224 307 Panda 129 157 Rhinoceros 225 306 Kangaroo 234 369
Table 2. Experimental environment configuration
View tableView in ArticleTable 2. Experimental environment configuration
名称 配置信息 操作系统 Ubuntu 18.04 GPU NVIDIA RTX 2080 Ti CPU Intel （R） Xeon （R） Platinum 8255 C CPU@2.50 GHz 深度学习框架 Pytorch-1.7.1 Python Python 3.8.17版本
Table 3. Experimental parameters configuration
View tableView in ArticleTable 3. Experimental parameters configuration
名称 配置信息 批次大小 4 优化器 SGD 动量参数 0.9 基类训练阶段学习率 0.005 微调阶段学习率 0.002 5
Table 4. Ablation experiments on the FSWA dataset
View tableView in ArticleTable 4. Ablation experiments on the FSWA dataset
方法 平衡L1损失 改进的PCB SA-RPN MSCE Novel AP50/% Novel AP/% 1 2 3 5 10 1 2 3 5 10 DeFRCN × × × × 59.3 72.2 74.0 78.4 81.0 29.7 39.6 43.0 45.2 48.4 Ours √ × × × 62.3 73.1 75.1 78.1 81.6 31.5 41.5 44.5 45.1 49.5 × √ × × 68.2 75.2 78.8 80.0 82.5 34.2 41.5 46.3 45.6 49.3 × × √ × 62.4 73.4 75.2 77.5 81.2 31.0 41.3 44.9 44.8 48.7 × × × √ 61.5 72.9 73.7 79.0 82.0 30.5 42.9 43.9 46.3 50.7 √ √ × × 69.2 75.3 78.6 80.4 82.4 35.4 42.1 46.3 45.4 48.8 √ √ √ × 68.4 76.6 78.8 80.9 82.7 33.9 42.4 45.7 46.2 48.7 √ √ √ √ 69.2 78.2 80.6 82.8 84.2 35.3 45.1 48.7 48.4 50.9
Table 5. Experimental results of different algorithms based on FSWA dataset
View tableView in ArticleTable 5. Experimental results of different algorithms based on FSWA dataset
Methods Novel AP50/% Novel AP/% 1-shot 2-shot 3-shot 5-shot 10-shot 1-shot 2-shot 3-shot 5-shot 10-shot TFA［ 13 ］29.2 31.2 33.5 39.3 46.5 13.5 19.5 20.6 23.3 27.7 FSDetView［ 25 ］13.5 25.6 38.3 55.0 64.6 5.4 9.5 21.1 27.6 35.0 FSCE［ 14 ］18.4 32.5 45.4 59.5 69.7 9.6 18.5 26.5 34.2 42.7 DeFRCN［ 15 ］59.3 72.2 74.0 78.4 81.0 29.7 39.6 43.0 45.2 48.4 VFA［ 26 ］39.8 55.8 62.5 70.0 72.5 15.5 21.5 25.8 29.0 36.1 MS-FSWD（Ours） 69.2 78.2 80.6 82.8 84.2 35.3 45.1 48.7 48.4 50.9
Table 6. Experimental results of different algorithms based on PASCAL VOC dataset
View tableView in ArticleTable 6. Experimental results of different algorithms based on PASCAL VOC dataset
Methods Novel Set 1 Novel Set 2 Novel Set 3 1-shot 2-shot 3-shot 5-shot 10-shot 1-shot 2-shot 3-shot 5-shot 10-shot 1-shot 2-shot 3-shot 5-shot 10-shot FSRW［ 10 ］14.2 23.6 29.8 36.5 35.6 12.3 19.6 25.1 31.4 29.8 12.5 21.3 26.8 33.8 31.0 TFA［ 13 ］25.3 36.4 42.1 47.9 52.8 18.3 27.5 30.9 34.1 39.5 17.9 27.2 34.3 40.8 45.6 FSDetView［ 25 ］24.2 35.3 42.2 49.1 57.4 21.6 24.6 31.9 37.0 45.7 21.2 30.0 37.2 43.8 49.6 FSCE［ 14 ］32.9 44.0 46.8 52.9 59.7 23.7 30.6 38.4 43.0 48.5 22.6 33.4 39.5 47.3 54.0 DeFRCN［ 15 ］40.2 53.6 58.2 63.6 66.5 29.5 39.7 43.4 48.1 52.8 35.0 38.3 52.9 57.7 60.8 FCT［ 27 ］38.5 49.6 53.5 59.8 64.3 25.9 34.2 40.1 44.9 47.4 34.7 43.9 49.3 53.1 56.3 VFA［ 26 ］47.4 54.4 58.5 64.5 66.5 33.7 38.2 43.5 48.3 52.4 43.8 48.9 53.3 58.1 60.0 MS-FSWD（Ours） 43.1 55.0 59.9 65.0 67.3 34.5 42.1 46.5 50.7 53.8 38.3 50.9 55.6 60.4 63.3
Get Citation
Copy Citation Text
Ke LIU, Shanling LIN, Xinyu SHI, Jianpu LIN, Shanhong LÜ, Zhixian LIN, Tailiang GUO. Few-shot wildlife detection based on multi-scale context extraction[J]. Chinese Journal of Liquid Crystals and Displays, 2025, 40(3): 516
Category:
Received: Jun. 11, 2024
Accepted: --
Published Online: Apr. 27, 2025
The Author Email: Zhixian LIN (lzx2005000@163.com)
