Myung-Joon Kwon

Hi again, I'm glad to hear that someone is implementing CAT-Net in Tensorflow 😃 I am releasing results for not using pretrained weights:  The results say that: 1)...

You should modify self.smallest to the smallest dataset size. In your case, set it as 400. The number 1869 is the size of IMD dataset.

Hi, Thanks for your report. It helps the forensic community to advance.

Hi, Heatmaps use a smaller size output, but you can use the `pred` variable in `tools/infer.py` to handle the full-resolution output. https://github.com/mjkwon2021/CAT-Net/blob/17d294d82f9e402484bf82637ba992fdc304c0e7/tools/infer.py#L135

@brightendavid Yes, the outputs have 1/4 resolution. This is due to two stride-2 convolutions at the very beginning, as explained in the paper 😊

Hi, I'm sorry to inform you that this code cannot run without CUDA. You'll see so many `.cuda()` in the whole project. If you wish to run it, you need...

Hi, You may use the masks in the `edgemask` directory in `4cam_splc` with a proper threshold (like 120/255). I've done this by: ``` mask = np.array(Image.open(mask_path))[:,:,1] mask[mask 0] = 1...

Sorry, I didn't create docker images. I think you should try installing jpegio and ask for help from their repo.

Hi, Thanks for your false-positive example report. Future works may focus more on images containing a relatively bright object compared to other regions, as your example suggests. Image manipulation detection...

Hi, This repo contains main training (segmentation) and not double JPEG pretraining (classification). I worked on another repo for classification but I just copy-pasted a double JPEG dataset here. I...