Home

Awesome

DQnet: Cross-Model Detail Querying for Camouflaged Object Detection

This is the official implementaion of paper DQnet: Cross-Model Detail Querying for Camouflaged Object Detection.

Illustration

With the contextual representation of ViT as global cues, our model queries crucial local details from the multi-scale CNN features. The enhanced representations have clear boundaries as well as few background noise, corresponding well with underlying camouflaged objects.

DQnet

Updates

Model

The model used pretrained weights for ResNet50 and ViT. They can be found at:

Usage

First clone the repository locally:

git clone https://github.com/CVPR23/DQnet.git

Second download the weights for ResNet50 and ViT.

Some core dependencies:

More details can be found in <./requirements.txt>

Datasets

More details can be found at:

For training:

You can use our default configuration, like this:

$ python main.py --model-name=DQnet --config=configs/DQnet/DQnet.py --datasets-info ./configs/_base_/dataset/dataset_configs.json --info demo

You can also use :

$ sh train.sh

For testing:

You can use our default configuration, like this:

$ python test.py 

You can also use :

$ sh test.sh

Paper Details

Method Detials

DQNet

RBQ

Comparison

Visualization of mutil-scale details querying