• [2024.05.28] 🔥 Release GIDAS dataset.
• [2024.12.02] 🎉 GASN accepted by Computers and Electronics in Agriculture. 🎉
• [2024.12.02] 🔥 Update Apache 2.0 License.
• [2025.01.16] 🔥 Release model, checkpoint and code. Update usage document.

• Dataset
• Release model, checkpoint and demo code.
• Usage document.

• Pytorch 1.13
• RTX 3060
• Windows 10 and Ubuntu 18.04

Images and density maps can be downloaded from here:

Google Drive and Baidu Netdisk

Backbone chexkpoint is in here:

We Use SAM Vit-H.

Use exp/vis_dataset.py to show annotations of GIDAS.

python ./exp/vis_dataset.py

Use exp/preprocess.py process dataset.

Firstly, we generate ground truth density map for GIDAS.

# generate GIDAS density map step 1, transform file type to mat
# it need only to run once
generate_density_mat()
# generate GIDAS density map step 2
# if set different sigma, it must be rerun
generate_density_map()

Secondly, we generate image imbedding for GIDAS to speed up calculations (adapt to 3060)

# generate GIDAS image embedding
gidas_preprocess()

Use exp/gasn_demo.py to run demo.

Select one of GIDAS to show effect, such as 103.jpg. (The 103.jpg is default.)

Open exp/dataset_gidas.py to set image in class GIDASDataset function __init__.

self.image_list = ["103.jpg"]

Then, you can run it.

python ./exp/gasn_demo.py

Some configurations is in exp/config.py.

Use exp/gasn_test.py to test gasn.

Set all of GIDAS to test.

Open exp/dataset_gidas.py to set image in class GIDASDataset function __init__.

self.image_list = os.listdir(self.image_dir)

Then, you can test it.

python exp/gasn_test.py

python exp/train.py