HVPNet

HVPNet: A Unified Bio-Inspired Network for General Salient and Camouflaged Object Detection

1. Model Selection Choose the appropriate model version based on your hardware capabilities and accuracy requirements:

Standard Version: HVPNet
Lightweight Version: HVPNet(-)

2. Task Selection Select the relevant model version according to your task and performance needs:

Standard Version: HVPNet (SMT-t + MobileNetV2) Lightweight Version: HVPNet(-) (MobileNetV2 + MobileNetV2)

3. Dataset Preparation We employed the following datasets for training the model:

RGB SOD: DUTS (Wang et al., 2017)

RGB-D SOD: NJUD (Ju et al., 2014), NLPR (Peng et al., 2014), DUTLF-Depth (Piao et al., 2019)

RGB-T SOD: VT5000 (Tu et al., 2022)

VSOD: DAVIS, FBMS, DAVSOD

RGB COD: COD10K (Fan et al., 2020), CAMO (Le et al., 2019)

RGB-D COD: COD10K, CAMO

VCOD: MoCA-Mask ├── dataset │ ├── RGB │ ├── Depth │ └── GT

Ensure that each dataset (RGB, Depth, GT) is placed in the corresponding folder.

4. Pretrained Weights Download the pretrained weights for your selected model:

SMT-t MobileNetV2

The pretrained weights can be found in the respective links provided in the documentation.

5. Training To train the model from scratch, execute the following command: python train_Net.py Ensure that your environment is properly set up with the necessary dependencies before training.

6. Testing For running inference and testing the model, use the following command:

RGB-SOD.pth：https://drive.google.com/file/d/1zjmgbNCOmKfMzWEPSr5jxjwl2stfTThO/view?usp=sharing

RGBD-SOD.pth:https://drive.google.com/file/d/1iZC3cSgpUCOtzmV_Ev5bWYxR9KaDQxnR/view?usp=sharing

RGBT-SOD.pth:https://drive.google.com/file/d/1ll0eFUw8s2haxRaUes214jL7dHNOgUxe/view?usp=sharing

RGBD-COD.pth：https://drive.google.com/file/d/1YVBNjWjOUgceqrPUYtNaGmVN7MQpOslX/view?usp=sharing

VCOD.pth：https://drive.google.com/file/d/1Xzrs_l6dQ1BELSWpHrzMVKKyG0kBRaDv/view?usp=sharing

python test_Net.py This will evaluate the trained model on the test set and generate results.