The model is built using PyTorch on an NVIDIA RTX A6000 GPU with a total memory size of 48 GB.

1. Introduction
2. Network
3. Data Preparation
4. Requirements Installation
5. Training
6. Evaluation
7. Results
8. Acknowledgement

Camouflaged object detection is an emerging and challenging computer vision task that requires identifying and segmenting objects that blend seamlessly into their environments due to high similarity in color, texture, and size. This task is further complicated by low-light conditions, partial occlusion, small object size, intricate background patterns, and the presence of multiple objects. While many sophisticated methods have been proposed for this task, current methods still struggle to precisely detect camouflaged objects in complex scenarios, especially with small and multiple objects, indicating room for improvement. We propose a Multi-Scale Recursive Network that extracts multi-scale features using a Pyramid Vision Transformer backbone and combines these features through specialized Attention-Based Scale Integration Units, allowing for selective feature merging. For more precise object detection, our decoder recursively refines features by incorporating Multi-Granularity Fusion Units. A novel recursive-feedback decoding strategy is developed to enhance global context understanding, helping the model overcome the challenges in this task. By jointly leveraging multi-scale learning and recursive feature optimization, our proposed method achieves performance gains, successfully detecting small and multiple camouflaged objects. Our model achieves state-of-the-art results on two benchmark camouflaged object detection datasets and ranks second on the remaining two.

This diagram illustrates the overall architecture of MSRNet.

In this research, we utilized four benchmark datasets for camouflaged object detection (CAMO, CHAMELEON, COD10K, and NC4K).

After downloading all datasets, you need to create a file named "dataset.yaml" and place it in the same directory as the main code folder.

The dataset.yaml file will include the paths for your Train and Test datasets. Please ensure that you place the datasets in the corresponding paths as you specified in the dataset.yaml file.

Your dataset.yaml file should look something like this:

# ICOD Datasets
cod10k_tr:
  {
    root: "YOUR_ROOT_DIRECTRY/ICOD_Datasets/Train/COD10K-TR",
    image: { path: "Image", suffix: ".jpg" },
    mask: { path: "Mask", suffix: ".png" },
  }
camo_tr:
  {
    root: "YOUR_ROOT_DIRECTRY/ICOD_Datasets/Train/CAMO-TR",
    image: { path: "Image", suffix: ".jpg" },
    mask: { path: "Mask", suffix: ".png" },
  }
cod10k_te:
  {
    root: "YOUR_ROOT_DIRECTRY/ICOD_Datasets/Test/COD10K-TE",
    image: { path: "Image", suffix: ".jpg" },
    mask: { path: "Mask", suffix: ".png" },
  }
camo_te:
  {
    root: "YOUR_ROOT_DIRECTRY/ICOD_Datasets/Test/CAMO-TE",
    image: { path: "Image", suffix: ".jpg" },
    mask: { path: "Mask", suffix: ".png" },
  }
chameleon:
  {
    root: "YOUR_ROOT_DIRECTRY/ICOD_Datasets/Test/CHAMELEON",
    image: { path: "Image", suffix: ".jpg" },
    mask: { path: "Mask", suffix: ".png" },
  }
nc4k:
  {
    root: "YOUR_ROOT_DIRECTRY/ICOD_Datasets/Test/NC4K",
    image: { path: "Imgs", suffix: ".jpg" },
    mask: { path: "GT", suffix: ".png" },
  }

• torch==2.1.2
• torchvision==0.16.2
• Others: pip install -r requirements.txt

python main_for_image.py --config configs/icod_train.py --pretrained --model-name EffB1_MSRNet
python main_for_image.py --config configs/icod_train.py --pretrained --model-name EffB4_MSRNet
python main_for_image.py --config configs/icod_train.py --pretrained --model-name PvtV2B2_MSRNet
python main_for_image.py --config configs/icod_train.py --pretrained --model-name PvtV2B3_MSRNet
python main_for_image.py --config configs/icod_train.py --pretrained --model-name PvtV2B4_MSRNet
python main_for_image.py --config configs/icod_train.py --pretrained --model-name PvtV2B5_MSRNet
python main_for_image.py --config configs/icod_train.py --pretrained --model-name RN50_MSRNet

python main_for_image.py --config configs/icod_train.py --model-name <MODEL_NAME> --evaluate --load-from <TRAINED_WEIGHT>

The prediction results of our highest-performing model (PvtV2B4_MSRNet) on CAMO, CHAMELEON, COD10K, and NC4K are available, along with the model weights.

A visual Comparison showing the superiority of MSRNet in detecting multiple (rows 1-3), small (rows 4 and 5), and tiny (rows 6 and 7) camouflaged objects.

This project builds upon the work of Lart Pang, incorporating key modifications to the decoding strategy and input scales to enhance its ability to detect small and multiple camouflaged objects.