CPGNet

KO | EN

gitlite — search

#python #react #ai #llm #golang #go #android #machine-learning #rust #linux #java #kubernetes #awesome-list #ios

(★ 10)

Official repository for CPGNet: Conditional Polarization Guidance for Camouflaged Object Detection.

CPGNet 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# CPGNet

Official anonymous implementation of **Conditional Polarization Guidance for Camouflaged Object Detection**.

## Overview

CPGNet is an asymmetric RGB-polarization framework for camouflaged object detection. Instead of treating polarization as a parallel modality for heavy fusion, CPGNet uses polarization as conditional structural guidance for hierarchical RGB representation learning.

The main components are:

- **PIM**: integrates AoLP and DoLP to produce reliable polarization guidance.
- **PGE**: injects polarization guidance into multi-scale RGB features.
- **EFM**: performs DoLP edge-guided frequency refinement.
- **IFD**: iteratively refines predictions through feedback decoding.

## Environment

```bash
conda create -n cpgnet python=3.8 -y
conda activate cpgnet
pip install -r requirements.txt
```

## Dataset Structure

Place the PCOD dataset under `datasets/PCOD`:

```text
datasets/PCOD/
  train/
    rgb/
    train-aop/
    train-dop/
    gt/
  test/
    rgb/
    test-aop/
    test-dop/
    gt/
```

The repository does not include datasets or prediction maps.

## Pretrained Backbone

Download the PVTv2-B2 pretrained weights and place them at:

```text
pretrained_pvt/pvt_v2_b2.pth
```

## Training

```bash
python train.py \
  --train_path datasets/PCOD/train \
  --test_path datasets/PCOD/test \
  --save_path ckpt \
  --batchsize 4 \
  --trainsize 704 \
  --epoch 100
```

## Testing

Download the trained CPGNet weight and place it at `ckpt/Net_epoch_best.pth`:

```text
Baidu Netdisk: https://pan.baidu.com/s/13Jf-GBwdYxahpQNdeJ8dkg
Extraction code: 8wpx
```

```bash
python test.py \
  --pth_path ckpt/Net_epoch_best.pth \
  --data_path datasets/PCOD/test \
  --save_path results/PCOD \
  --testsize 704
```

## Evaluation

```bash
python evaluation/evaluate.py
```

## Reproducibility

The default training configuration follows the manuscript: input size 704, batch size 4, AdamW optimizer, and 100 training epochs.

CPGNet

CPGNet 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

CPGNet

CPGNet 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README