Change is Everywhere: Single-Temporal Supervised Object Change Detection in Remote Sensing Imagery (ICCV 2021)

Last update: Dec 13, 2022

Overview

Change is Everywhere
Single-Temporal Supervised Object Change Detection
in Remote Sensing Imagery

by Zhuo Zheng, Ailong Ma, Liangpei Zhang and Yanfei Zhong

[Paper] [BibTeX]

This is an official implementation of STAR and ChangeStar in our ICCV 2021 paper Change is Everywhere: Single-Temporal Supervised Object Change Detection for High Spatial Resolution Remote Sensing Imagery.

We hope that STAR will serve as a solid baseline and help ease future research in weakly-supervised object change detection.

News

2021/08/28, The code is available.
2021/07/23, The code will be released soon.
2021/07/23, This paper is accepted by ICCV 2021.

Features

Learning a good change detector from single-temporal supervision.
Strong baselines for bitemporal and single-temporal supervised change detection.
A clean codebase for weakly-supervised change detection.
Support both bitemporal and single-temporal supervised settings

Citation

If you use STAR or ChangeStar (FarSeg) in your research, please cite the following paper:

@inproceedings{zheng2021change,
  title={Change is Everywhere: Single-Temporal Supervised Object Change Detection for High Spatial Resolution Remote Sensing Imagery},
  author={Zheng, Zhuo and Ma, Ailong and Liangpei Zhang and Zhong, Yanfei},
  booktitle={Proceedings of the IEEE international conference on computer vision},
  pages={},
  year={2021}
}

@inproceedings{zheng2020foreground,
  title={Foreground-Aware Relation Network for Geospatial Object Segmentation in High Spatial Resolution Remote Sensing Imagery},
  author={Zheng, Zhuo and Zhong, Yanfei and Wang, Junjue and Ma, Ailong},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={4096--4105},
  year={2020}
}

Getting Started

Install EVer

pip install --upgrade git+https://github.com/Z-Zheng/ever.git

Requirements:

pytorch >= 1.6.0
python >=3.6

Prepare Dataset

Download xView2 dataset (training set and tier3 set) and LEVIR-CD dataset.
Create soft link

ln -s </path/to/xView2> ./xView2
ln -s </path/to/LEVIR-CD> ./LEVIR-CD

Training and Evaluation under Single-Temporal Supervision

bash ./scripts/trainxView2/r50_farseg_changemixin_symmetry.sh

Training and Evaluation under Bitemporal Supervision

bash ./scripts/bisup_levircd/r50_farseg_changemixin.sh

License

ChangeStar is released under the Apache License 2.0.

Comments

Can ChangeStar be used for general CD?

hi,

Thanks for the great work. I wonder, can this work be used for general change detection? i.e., multi-class not just single class.

If yes, do you have done the experiments? Thanks!

opened by Richardych 3
hello, how to add changemixin when use bitemporal supervised
hello I have question about your repo:

how to add changeminxin when use bitemporal supervised, i see it in your paper table 4 but i cant find in codes?

could changestar use LEVIR-CD train Single-Temporal(another dataset is too big for train, i cant download it)

are your bitemporal suprvised methods just use torch.cat in the final layer? sorry for ask these question,
opened by csliuchang 3
ValueError: Requested crop size (512, 512) is larger than the image size (384, 384)

Traceback (most recent call last): File "./train_sup_change.py", line 48, in blob = trainer.run(after_construct_launcher_callbacks=[register_evaluate_fn]) File "/home/yujianzhi/anaconda3/envs/CStar/lib/python3.7/site-packages/ever/api/trainer/th_amp_ddp_trainer.py", line 117, in run test_data_loader=kw_dataloader['testdata_loader']) File "/home/yujianzhi/anaconda3/envs/CStar/lib/python3.7/site-packages/ever/core/launcher.py", line 232, in train_by_config signal_loss_dict = self.train_iters(train_data_loader, test_data_loader=test_data_loader, **config) File "/home/yujianzhi/anaconda3/envs/CStar/lib/python3.7/site-packages/ever/core/launcher.py", line 174, in train_iters is_master=self._master) File "/home/yujianzhi/anaconda3/envs/CStar/lib/python3.7/site-packages/ever/core/iterator.py", line 30, in next data = next(self._iterator) File "/home/yujianzhi/anaconda3/envs/CStar/lib/python3.7/site-packages/torch/utils/data/dataloader.py", line 435, in next data = self._next_data() File "/home/yujianzhi/anaconda3/envs/CStar/lib/python3.7/site-packages/torch/utils/data/dataloader.py", line 475, in _next_data data = self._dataset_fetcher.fetch(index) # may raise StopIteration File "/home/yujianzhi/anaconda3/envs/CStar/lib/python3.7/site-packages/torch/utils/data/_utils/fetch.py", line 44, in fetch data = [self.dataset[idx] for idx in possibly_batched_index] File "/home/yujianzhi/anaconda3/envs/CStar/lib/python3.7/site-packages/torch/utils/data/_utils/fetch.py", line 44, in data = [self.dataset[idx] for idx in possibly_batched_index] File "/home/yujianzhi/anaconda3/envs/CStar/lib/python3.7/site-packages/torch/utils/data/dataset.py", line 218, in getitem return self.datasets[dataset_idx][sample_idx] File "/home/yujianzhi/tem/ChangeStar-master/data/levir_cd/dataset.py", line 30, in getitem blob = self.transforms(**dict(image=imgs, mask=gt)) File "/home/yujianzhi/anaconda3/envs/CStar/lib/python3.7/site-packages/albumentations/core/composition.py", line 191, in call data = t(force_apply=force_apply, **data) File "/home/yujianzhi/anaconda3/envs/CStar/lib/python3.7/site-packages/albumentations/core/transforms_interface.py", line 90, in call return self.apply_with_params(params, **kwargs) File "/home/yujianzhi/anaconda3/envs/CStar/lib/python3.7/site-packages/albumentations/core/transforms_interface.py", line 103, in apply_with_params res[key] = target_function(arg, **dict(params, **target_dependencies)) File "/home/yujianzhi/anaconda3/envs/CStar/lib/python3.7/site-packages/albumentations/augmentations/crops/transforms.py", line 48, in apply return F.random_crop(img, self.height, self.width, h_start, w_start) File "/home/yujianzhi/anaconda3/envs/CStar/lib/python3.7/site-packages/albumentations/augmentations/crops/functional.py", line 28, in random_crop crop_height=crop_height, crop_width=crop_width, height=height, width=width ValueError: Requested crop size (512, 512) is larger than the image size (384, 384) Traceback (most recent call last): File "/home/yujianzhi/anaconda3/envs/CStar/lib/python3.7/runpy.py", line 193, in _run_module_as_main "main", mod_spec) File "/home/yujianzhi/anaconda3/envs/CStar/lib/python3.7/runpy.py", line 85, in _run_code exec(code, run_globals) File "/home/yujianzhi/anaconda3/envs/CStar/lib/python3.7/site-packages/torch/distributed/launch.py", line 260, in main() File "/home/yujianzhi/anaconda3/envs/CStar/lib/python3.7/site-packages/torch/distributed/launch.py", line 256, in main cmd=cmd) subprocess.CalledProcessError: Command '['/home/yujianzhi/anaconda3/envs/CStar/bin/python', '-u', './train_sup_change.py', '--local_rank=0', '--config_path=levircd.r50_farseg_changestar_bisup', '--model_dir=./log/bisup-LEVIRCD/r50_farseg_changestar']' returned non-zero exit status 1.

it says： ValueError: Requested crop size (512, 512) is larger than the image size (384, 384) but my img is 512*512 exactly.

opened by themoongodyue 3
How to get the bitemporal images' labels if the model is trained on LEVIR-CD dataset?

Hello, I'm very interested in your work, but I encountered a problem in the process of research. If the model is trained on the LEVIR-CD dataset, how to obtain the changed labels when there are no segmentation maps for each bitemporal image in the dataset? I would appreciate it if you could solve my problems.

opened by SONGLEI-arch 2
Reproduction Problem

Hello author.

Your work is great!

But I ran into a problem while running your code.

The performance came as shown in the picture below, but this number is much higher than the number in table1 of your paper. (IoU) Can you tell me the reason?

All hyperparameters and data are identical.

opened by seominseok0429 1
AssertionError error

Hello, this is really great work. I have one question for you. The LEVIR-CD dataset trains well, but the xview2 dataset gives the following unknown error.

Do you have any idea how to fix it? All processes follow the recipe exactly .

opened by seominseok0429 1
RuntimeError: NCCL error in: /pytorch/torch/lib/c10d/ProcessGroupNCCL.cpp:911, unhandled system error, NCCL version 2.7.8

i have crazy,help me please

Traceback (most recent call last): File "./train_sup_change.py", line 48, in blob = trainer.run(after_construct_launcher_callbacks=[register_evaluate_fn]) File "/home/cy/miniconda3/envs/STAnet/lib/python3.8/site-packages/ever/api/trainer/th_amp_ddp_trainer.py", line 98, in run kwargs.update(dict(model=self.make_model())) File "/home/cy/miniconda3/envs/STAnet/lib/python3.8/site-packages/ever/api/trainer/th_amp_ddp_trainer.py", line 87, in make_model model = nn.parallel.DistributedDataParallel( File "/home/cy/miniconda3/envs/STAnet/lib/python3.8/site-packages/torch/nn/parallel/distributed.py", line 496, in init dist._verify_model_across_ranks(self.process_group, parameters) RuntimeError: NCCL error in: /pytorch/torch/lib/c10d/ProcessGroupNCCL.cpp:911, unhandled system error, NCCL version 2.7.8 ncclSystemError: System call (socket, malloc, munmap, etc) failed. ERROR:torch.distributed.elastic.multiprocessing.api:failed (exitcode: 1) local_rank: 0 (pid: 31335) of binary: /home/cy/miniconda3/envs/STAnet/bin/python ERROR:torch.distributed.elastic.agent.server.local_elastic_agent:[default] Worker group failed

opened by themoongodyue 1
Evaluation

Excuse me, I want to know how this module behave inference after training the model. And if you can offer an link for usage of 'ever' Lib, that will be fantastic

opened by LIUZIJING-CHN 1
changestar_sisup results

Hi, I have trained the model under single-temporal supervision, but the F1 result is only 0.73，which is worse than the result in your paper. Is there anything wrong with my experiment, below is my training log：

1666753326.225779.log

After training I only test the LEVIR-CD test set.

opened by max2857 0
A question about PCC

Hello,I have a question about PCC:

PCC is mentioned in the paper. After obtaining the classification result through the segmentation model, how to obtain the change detection result through the classification result? Is it a direct subtraction?

opened by Hyd1999618 0
[Feature] support [0~255] gt

The original dataset of LEVIR-CD consists of 0 and 255.

However, the segmentation loss of this code works only when it consists of 0 and 1.

Therefore, I added a code to change gt's 255 to 1.

opened by seominseok0429 1

Releases(v0.1.0)

v0.1.0(Nov 4, 2021)

The target files of xView2 tier3 should be generated from the given geojson files. For convenience, we release the generated target files here.
Source code(tar.gz)
Source code(zip)
xView2-tier3-targets.zip(22.63 MB)

Owner

Zhuo Zheng

CV IN RS. Ph.D. Student.

GitHub Repository

1st place solution to the Satellite Image Change Detection Challenge hosted by SenseTime

209 Jan 01, 2023

VR Viewport Pose Model for Quantifying and Exploiting Frame Correlations

This repository contains the introduction to the collected VRViewportPose dataset and the code for the IEEE INFOCOM 2022 paper: "VR Viewport Pose Model for Quantifying and Exploiting Frame Correlatio

0 Aug 10, 2022

tree-math: mathematical operations for JAX pytrees

tree-math: mathematical operations for JAX pytrees tree-math makes it easy to implement numerical algorithms that work on JAX pytrees, such as iterati

137 Dec 28, 2022

Realtime segmentation with ENet, the fast and accurate segmentation net.

Enet This is a realtime segmentation net with almost 22 fps on GTX1080 ti, and the model size is very small with only 28M. This repo contains the infe

14 Aug 30, 2022

Unsupervised phone and word segmentation using dynamic programming on self-supervised VQ features.

Unsupervised Phone and Word Segmentation using Vector-Quantized Neural Networks Overview Unsupervised phone and word segmentation on speech data is pe

13 Dec 11, 2022

Editing a Conditional Radiance Field

Editing Conditional Radiance Fields Project | Paper | Video | Demo Editing Conditional Radiance Fields Steven Liu, Xiuming Zhang, Zhoutong Zhang, Rich

216 Dec 30, 2022

A spatial genome aligner for analyzing multiplexed DNA-FISH imaging data.

jie jie is a spatial genome aligner. This package parses true chromatin imaging signal from noise by aligning signals to a reference DNA polymer model

9 Sep 29, 2022

Use CLIP to represent video for Retrieval Task

A Straightforward Framework For Video Retrieval Using CLIP This repository contains the basic code for feature extraction and replication of results.

54 Dec 22, 2022

pixelNeRF: Neural Radiance Fields from One or Few Images

pixelNeRF: Neural Radiance Fields from One or Few Images Alex Yu, Vickie Ye, Matthew Tancik, Angjoo Kanazawa UC Berkeley arXiv: http://arxiv.org/abs/2

1k Jan 04, 2023

《LXMERT: Learning Cross-Modality Encoder Representations from Transformers》(EMNLP 2020)

The Most Important Thing. Our code is developed based on: LXMERT: Learning Cross-Modality Encoder Representations from Transformers

53 Dec 16, 2022

CC-GENERATOR - A python script for generating CC

CC-GENERATOR A python script for generating CC NOTE: This tool is for Educationa

6 Oct 14, 2022

Interactive dimensionality reduction for large datasets

BlosSOM 🌼 BlosSOM is a graphical environment for running semi-supervised dimensionality reduction with EmbedSOM. You can use it to explore multidimen

19 Dec 14, 2022

Streaming Anomaly Detection Framework in Python (Outlier Detection for Streaming Data)

Python Streaming Anomaly Detection (PySAD) PySAD is an open-source python framework for anomaly detection on streaming multivariate data. Documentatio

181 Dec 18, 2022

A copy of Ares that costs 30 fucking dollars.

Finalement, j'ai décidé d'abandonner cette idée, je me suis comporté comme un enfant qui été en colère. Comme m'ont dit certaines personnes j'ai des c

24 Apr 14, 2022

Python script that analyses the given datasets and comes up with the best polynomial regression representation with the smallest polynomial degree possible

Python script that analyses the given datasets and comes up with the best polynomial regression representation with the smallest polynomial degree possible, to be the most reliable with the least com

2 Aug 01, 2022