MultiBranch_CNNCopyMove_Disambiguation

KO | EN

gitlite — search

#python #llm #golang #android #machine-learning #linux #docker #awesome #kubernetes #awesome-list #pytorch #security #openai #open-source

MultiBranch_CNNCopyMove_Disambiguation

(★ 14)

Implementation of "Copy Move Source-Target Disambiguation through Multi-Branch CNNs" by Mauro Barni, Quoc-Tin Phan, Benedetta Tondi

MultiBranch_CNNCopyMove_Disambiguation 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

![Image](https://raw.githubusercontent.com/andreacos/MultiBranch_CNNCopyMove_Disambiguation/master/resources/vippdiism.png)

# Multi-Branch CNN Copy Move Source-Target Disambiguation

This repository is an implementation of the method in

"Copy Move Source-Target Disambiguation through Multi-Branch CNNs" 
Mauro Barni, Quoc-Tin Phan, Benedetta Tondi

The disambiguation method allows to identify source and target region of a copy-move by means of a multi-branch 
CNN-based architecture, capable to learn suitable features by looking at the presence of interpolation  artifacts 
and boundary inconsistencies.

Code implemented by Quoc-Tin Phan (dimmoon2511[at]gmail.com)

## Step 1. Prerequisites

All prerequisites should be listed in requirements.txt.

To avoid conflicts with your system, it is encouraged to first create virtual environment in Anaconda.

Install all prerequisites:

```pip install -r resources/requirements.txt```

## Step 2. Download pre-trained models

Download [pre-trained models](https://drive.google.com/file/d/1LS2Z8bLnL_non7Ibcl9QWd1zBGl_PAD6)
and unzip the folder.

## Step 3 (optional). Training

To generate synthetic datasets use the Matlab scripts in ./resources/db_gen:

* For training 4-Twins net: ```run_create_scribble_db.m```.

* For training Siamese net: ```run_create_scribble_rigid_cm_db.m```

**Added on 17/05/2021.** The list of images used to train the networks is now available [here](https://github.com/andreacos/MultiBranch_CNNCopyMove_Disambiguation/blob/master/resources/db_gen/TRAINING_SET.txt). To use it, proceed as follows:  
* download the three required datasets: RAISE2K, VISION and Dresden from their respective websites; 
* use the list to move only the training images into a single folder; 
* edit the [Matlab db generation code](https://github.com/andreacos/MultiBranch_CNNCopyMove_Disambiguation/blob/master/resources/db_gen/run_create_scribble_rigid_cm_db.m) to point to that folder (e.g. this line of code for the rigid case: `input_path  = '../datasets/RAISE_DRESDEN_VISION_mixed')`

To train 4-Twins net, re-configure parameters in **train\_4twins.py**:

* ```pos_neg.txt```: path to the file whose each line contains absolute path to a training image.
* ```max_items```: the number of training images (the number of lines in ```pos_neg.txt``` should be equal or over this number).
* ```use_gpu```: the list of GPU's ID that we use. A batch of data is equally split and distributed on each GPU. The final loss is synthesized from each GPU.
* ```working_dir```: where to save the model and training logs.
* ```pretrained_resnet```: directory of resnet50 pretrained on ImageNet (you can download from [here](http://download.tensorflow.org/models/resnet_v2_50_2017_04_14.tar.gz), and decompress into this directory).

Similarly, re-configure parameters in **train\_siamese.py** to train Siamese net:

* ```pos_lst.txt```: path to the file whose each line contains absolute path to a training image. 
* Other parameters can be configured analogously to 4-Twins net.

## Step 4. Prediction

Execute:

```python predict_4twins.py```

```python predict_siamese.py```

```python predict_fusion.py```

MultiBranch_CNNCopyMove_Disambiguation

MultiBranch_CNNCopyMove_Disambiguation 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

MultiBranch_CNNCopyMove_Disambiguation

MultiBranch_CNNCopyMove_Disambiguation 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README