dklive1999 / maps-net

Paper code: Image tampering localization network based on multi-class attention and progressive subtraction

MAPS-Net

Paper code: Image tampering localization network based on multi-class attention and progressive subtraction

Environment

• Ubuntu18.04
• Python 3.8
• PyTorch 1.9.0 + Cuda 11.1
• Detail python librarys can found in requirements.txt

Dataset

An example of the dataset index file is given as datasets/Casiav2.txt, where each line contains:

 img_path mask_path label

• 0 represents the authentic and 1 represents the manipulated.
• For an authentic image, the mask_path is "None".
• For wild images without mask groundtruth, the index should at least contain "img_path" per line.

For example, each line in txt file should like this:`

• Authentic image:

  ./Casiav2/authentic/Au_ani_00001.jpg None None 0
  

• Manipulated image with pre-generated edge mask:

  ./Casiav2/tampered/Tp_D_CND_M_N_ani00018_sec00096_00138.tif ./Casiav2/mask/Tp_D_CND_M_N_ani00018_sec00096_00138_gt.png ./Casiav2/edge/Tp_D_CND_M_N_ani00018_sec00096_00138_gt.png 1
  

• Manipulated image without pre-generated edge mask:

  ./Casiav2/tampered/Tp_D_CND_M_N_ani00018_sec00096_00138.tif ./Casiav2/mask/Tp_D_CND_M_N_ani00018_sec00096_00138_gt.png None 1
  

• You should follow the format and generate your own "path file" in a xxxx.txt.

Limits: At this time, the edge mask can only be generated during training and cannot be pre generated. This will be a little bit slow. Since every Epoch you will generate a edge mask for each image, however, they are always the same edge mask. Better choice should be generate the edge mask from the ground truth mask before start training. Script for pre-generate the edge mask will release later.

Training sets

• CASIAv2

Test sets

• CASIAv1
• Columbia
• COVERAGE
• nist16

Training

• Please set the train image path in train_base.py, then run train_lanch.py with Python.
• In the models folder we have provided several working python files, namely "maps_net-xxx.py", please select the .py file you need to run and rename it to "maps_net.py " for proper training. These .py files were also used by us in the ablation experiments.

Evaluation

• Please set the test image path in inference.py and run inference.py with Python, then run evaluate.py with Python.

maps-net's Issues

Source of the paper

Thanks for your coding work. The advantages of this model can be seen below. According to the title of the paper you provided, no relevant information was found. I really want to read this paper. Could you please ask the author to provide the source of the paper.