Awesome

DSALVANet

This is the code for Few-Shot Object Counting with Dynamic Similarity-Aware in Latent Space.

<div align=center> <img src="./test_data/4297.jpg" width="300"/><img src="./output/output.jpg" width="300"/> </div> <div align=center> <img src="./test_data/P0787.png" width="250"/><img src="./output/output1.jpg" width="250"/><img src="./output/output2.jpg" width="250"/> </div>

Dependencies

We are good in the environment:

python $\geqslant$ 3.7
numpy 1.23.5
torch 2.0.1
torchvision 0.15.2
opencv 3.4.10

Installation

To install the required packages, please run:

pip install -r requirements.txt

Pre-trained models

To test the code, you can download the FSC147 pre-trained model of DSALVANet: checkpoint_200.pth

Usage

We provide the test code for our model.
Modify the path of input data and pre-trained model for testing.

python test.py -w ./checkpoints/checkpoint_200.pth -i ./test_data/4297.jpg -b ./test_data/4297.txt

Finally, the model will output a result with count values and visualization in './output' folder.

Others

We will release more details of DSALVANet after the paper is officially publiced in the journal.

Before that, if you have any questions, you can contact me via email: kadvinj@outlook.com.