MICPL: Motion-Inspired Cross-Pattern Learning for Small Object Detection in Satellite Videos

Introduction

For small object detection, vision pattern can only provide a limited support to feature learning. Most prior schemes mainly depend on a single vision pattern to learn object features, seldom considering more latent motion patterns. In real world, humans often efficiently perceive small objects through multi-pattern signals. Inspired by this observation, this paper attempts to address small object detection from a new prospective of latent pattern learning. To fulfill this purpose, it regards a real-world moving object as the spatio-temporal sequences of a static object to capture latent motion patterns. In view of this, we propose a Motion-Inspired Cross-Pattern Learning (MICPL) scheme to capture the motion patterns for moving small object scenarios.

Datasets

• Satellite videos are available at DSFNet or VISO.
• DroneCrowd dataset and baseline model: download from DroneCrowd.

Prerequisite

• Tested on Ubuntu 20.04, with Python 3.10, PyTorch 1.12, Torchvision 0.13.1, CUDA 11.3, and 2x NVIDIA 3080.
• You can follow CenterNet to build the conda environment but remember to replace the DCNv2 used here with the used DCNv2 by CenterNet (Because we used the latested version of DCNv2).
• You can also follow CenterNet and DSFNet to build the conda environment with Python 3.7, PyTorch 1.7, Torchvision 0.8.1, CUDA 10.2 and run this code.

Usage

Train

python train.py --model_name MICPL \
		--gpus 0,1 \
		--lr 1.25e-4 \
		--lr_step 30,45 \
		--num_epochs 55 \
		--batch_size 2 \
		--val_intervals 5 \ 
		--test_large_size True \
		--datasetname  {dataset name} \
		--load_model ./lib/model/pre-trained.pth \
		--data_dir {dataset path}

Test and visulization

python test.py --model_name MICPL \
               --gpus 0 \
               --load_model {checkpoint path} \
               --test_large_size True \
               --datasetname {dataset name} \
               --data_dir {dataset path}

Evaluation

python testSaveMat.py

python evaluation.py

Contact

If any questions, kindly contact with Shengjia Chen via e-mail: csj_uestc@126.com.

References

1. Wen, Longyin, et al. "Detection, tracking, and counting meets drones in crowds: A benchmark." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021.
2. X. Zhou, D. Wang, and P. Krahenbuhl, "Objects as points," arXiv preprint arXiv:1904.07850, 2019.
3. Xiao, Chao, et al. "DSFNet: Dynamic and static fusion network for moving object detection in satellite videos." IEEE Geoscience and Remote Sensing Letters 19 (2021): 1-5.