LSKNet

---

Jittor implementation at github.com/NK-JittorCV/nk-remote

---

Update[8/1/2025] Supports Strip R-CNN: Large Strip Convolution for Remote Sensing Object Detection. Official code is available at YXB-NKU/Strip-R-CNN

We have added the config for our latest work, Strip-R-CNN arxiv. In this paper ,we have reached 82.75% mAP on DOTA1.0 dataset, setting a new state-of-the-art record.

This repository is the official implementation of IJCV (accepted in 2024) "LSKNet: A Foundation Lightweight Backbone for Remote Sensing" at: IJCV or arxiv

Our conference version: ICCV 2023 "Large Selective Kernel Network for Remote Sensing Object Detection" at: ICCV Open Access

Abstract

Recent research on remote sensing object detection has largely focused on improving the representation of oriented bounding boxes but has overlooked the unique prior knowledge presented in remote sensing scenarios. Such prior knowledge can be useful because tiny remote sensing objects may be mistakenly detected without referencing a sufficiently long-range context, and the long-range context required by different types of objects can vary. In this paper, we take these priors into account and propose the Large Selective Kernel Network (LSKNet). LSKNet can dynamically adjust its large spatial receptive field to better model the ranging context of various objects in remote sensing scenarios. To the best of our knowledge, this is the first time that large and selective kernel mechanisms have been explored in the field of remote sensing object detection. Without bells and whistles, our lightweight LSKNet sets new state-of-the-art scores on standard remote sensing classification, object detection and semantic segmentation benchmarks. Based on a similar technique, we rank 2nd place in 2022 the Greater Bay Area International Algorithm Competition

Introduction

This repository is the official implementation of IJCV 2024 "LSKNet: A Foundation Lightweight Backbone for Remote Sensing" at: arxiv

The master branch is built on MMRotate which works with PyTorch 1.6+.

LSKNet backbone code is placed under mmrotate/models/backbones/, and the train/test configure files are placed under configs/lsknet/

Results and models

Imagenet 300-epoch pre-trained LSKNet-T backbone: Download

Imagenet 300-epoch pre-trained LSKNet-S backbone: Download

Imagenet 300-epoch pre-trained Strip R-CNN-T backbone: Download

Imagenet 300-epoch pre-trained Strip R-CNN-S backbone: Download

DOTA1.0

| Model | mAP | Angle | lr schd | Batch Size | Configs | Download | note | | :--------------------------------------------------------: | :---: | :---: | :-----: | :--------: | :--------------------------------------------------------------------------: | :----------------------------------------------------------------------------------------------------------------------------------: | :----------: | | RTMDet-l (1024,1024,-) | 81.33 | - | 3x-ema | 8 | - | - | Prev. Best | | LSKNet_T (1024,1024,200) + ORCNN | 81.37 | le90 | 1x | 2*8 | lsk_t_fpn_1x_dota_le90 | model | log | | | LSKNet_S (1024,1024,200) + ORCNN | 81.64 | le90 | 1x | 1*8 | lsk_s_fpn_1x_dota_le90 | model | log | | | LSKNet_S* (1024,1024,200) + ORCNN | 81.85 | le90 | 1x | 1*8 | lsk_s_ema_fpn_1x_dota_le90 | model | log | EMA Finetune | | LSKNet_S (1024,1024,200) + Roi_Trans | 81.22 | le90 | 1x | 2*8 | lsk_s_roitrans_fpn_1x_dota | model | log | | | LSKNet_S (1024,1024,200) + R3Det | 80.08 | oc | 1x | 2*8 | lsk_s_r3det_fpn_1x_dota | model | log | | | LSKNet_S (1024,1024,200) + S2ANet | 81.32 | le135 | 1x | 2*