donaldlee2008 / tiny_model_4_cd

Official implementation of TINYCD: A (NOT SO) DEEP LEARNING MODEL FOR CHANGE DETECTION

Tiny_CD

This is the implementation of: TINYCD: A (NOT SO) DEEP LEARNING MODEL FOR CHANGE DETECTION

You can find the complete work here: https://arxiv.org/abs/2207.13159

Results

samples_reduced.mp4

In the following table we report the quantitative comparison of our model and other State-of-the-art models. F1 scores are reported in percentage, parameters in Millions and Flops in Gigaflops

Here we report a visual comparison between the mask generated by TinyCD and BIT. We selected negative samples in the fairness way in order to show both similarities and differences between the obtained results.

Finally, we show a complete sequence of intermediate masks and final binary classification mask to show the ability of our tiny model in detecting detailed objects.

Installation and Requirements

The easiest way to reproduce our results is to create a folder named "TinyCD" on your device and then clone this repository in "TinyCD":

git clone https://github.com/AndreaCodegoni/Tiny_model_4_CD.git

Then, you can create a virtual conda environment named TinyCD with the following cmd:

conda create --name TinyCD --file requirements.txt
conda activate TinyCD

Dataset

You can find the original datasets at these two links:

LEVIR-CD: https://justchenhao.github.io/LEVIR/

WHU-CD: https://study.rsgis.whu.edu.cn/pages/download/building_dataset.html

Then, for each dataset, you have to organise the data in the following way:

A: images of t1 phase;

B: images of t2 phase;

label: label maps;

list: contains train.txt, val.txt and test.txt, each file records the image names (XXX.png) in the change detection dataset.

If you prefer, you can download the pre-processed dataset using the following:

LEVIR-CD

wget https://www.dropbox.com/s/h9jl2ygznsaeg5d/LEVIR-CD-256.zip

WHU-CD

wget https://www.dropbox.com/s/r76a00jcxp5d3hl/WHU-CD-256.zip

If you have any trouble with the datasets, feel free to contact us.

Evaluate pretrained models

If you want to evaluate your trained model, or if you want to reproduce the paper results with the pretrained models that you can find in the "pretrained_models" folder, you can run:

python test_ondata.py --datapath "Your_data_path" --modelpath "Your_path_to_pretrained_model"

Train your model

You can re-train our model, or if you prefer you can play with the parameters of our model and then train it using

python training.py --datapath "Your_data_path" --log-path "Path_to_save_logs_and_models_checkpoints"

References

We want to mention the following repositories that greatly help us in our work:

• https://github.com/justchenhao/BIT_CD We have used this repository in the visual comparison and to report other state-of-the-art results on the two datasets.
• https://github.com/wgcban/ChangeFormer/ and https://github.com/wgcban/SemiCD for the datasets.

License

Code is released for non-commercial and research purposes only. For commercial purposes, please contact the authors.