vinthony / deep-blind-watermark-removal

[AAAI 2021] Split then Refine: Stacked Attention-guided ResUNets for Blind Single Image Visible Watermark Removal

This repo contains the code and results of the AAAI 2021 paper:

Split then Refine: Stacked Attention-guided ResUNets for Blind Single Image Visible Watermark Removal
Xiaodong Cun, Chi-Man Pun^*
University of Macau

Datasets | Models | Paper | 🔥Online Demo!(Google CoLab)

---

The overview of the proposed two-stage framework. Firstly, we propose a multi-task network, SplitNet, for watermark detection, removal, and recovery. Then, we propose the RefineNet to smooth the learned region with the predicted mask and the recovered background from the previous stage. As a consequence, our network can be trained in an end-to-end fashion without any manual intervention. Note that, for clarity, we do not show any skip-connections between all the encoders and decoders.

---

The whole project will be released in the January of 2021 (almost).

Datasets

We synthesized four different datasets for training and testing, you can download the dataset via huggingface.

Pre-trained Models

• 27kpng_model_best.pth.tar (google drive)

Other Pre-trained Models are still reorganizing and uploading, it will be released soon.

Demos

An easy-to-use online demo can be founded in google colab.

The local demo will be released soon.

Pre-requirements

pip install -r requirements.txt

Train

Besides training our methods, here, we also give an example of how to train the s2am under our framework. More details can be found in the shell scripts.

bash examples/evaluation.sh

Test

bash examples/test.sh

Acknowledgements

The author would like to thanks Nan Chen for her helpful discussion.

Part of the code is based upon our previous work on image harmonization s2am

Citation

If you find our work useful in your research, please consider citing:

@misc{cun2020split,
      title={Split then Refine: Stacked Attention-guided ResUNets for Blind Single Image Visible Watermark Removal}, 
      author={Xiaodong Cun and Chi-Man Pun},
      year={2020},
      eprint={2012.07007},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Contact

Please contact me if there is any question (Xiaodong Cun [email protected])

deep-blind-watermark-removal's Issues

How to generate "mask of watermark" when i want to use my own dataset to train?

Can I get some evaluating metric or something else?

Excuse me , can I know how to evaluate the result of removing watermarks?

Do you have some metric like mAP or IoU? I just wonder how to evaluate the results of removing watermarks.

Can I get some evaluating metric or something else?

Thx!

Corrupted images in dataset

Thanks for your excellent work! However, I found some corrupted images in the dataset. In 10kgray, when reading val_images/image/COCO_val2014_000000087481-Kaporal_Jeans_Logo-185.png, train_images/image/COCO_val2014_000000260525-Breyers_Logo-163.png, train_images/image/COCO_val2014_000000510651-University_Of_Oxford_Logo_Text-179.png using imageio.v2.imread, an error image file is truncated occurs. In fact the latter two images do not contain the watermark and are thus not usable.

The three truncated images.

Modifying COCO dataset class to include 256x256 Random Crop gives CUDA error

Hi, I was trying to train the model on the Logo Dataset that you had provided, without resizing the Images.
Instead of Hard resize to 256x256 which distorts the aspect ratio, I decided to use random cropping however after a few epochs I'd run into CUDA error

Here is the error:

/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [0,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [1,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [2,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [3,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [4,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [5,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [6,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [7,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [8,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [9,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [10,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [11,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [12,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [13,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [14,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [15,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [16,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [17,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [18,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [19,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [20,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [21,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [22,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [23,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [24,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [25,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [26,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [27,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [28,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [29,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [30,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [228,0,0], thread: [31,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [32,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [33,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [34,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [35,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [36,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [37,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [38,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [39,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [40,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [41,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [42,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [43,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [44,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [45,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [46,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [47,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [48,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [49,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [50,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [51,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [52,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [53,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [54,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [55,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [56,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [57,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [58,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [59,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [60,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [61,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [62,0,0] Assertion input_val >= zero && input_val <= one failed.
/pytorch/aten/src/ATen/native/cuda/Loss.cu:102: operator(): block: [197,0,0], thread: [63,0,0] Assertion input_val >= zero && input_val <= one failed.
Traceback (most recent call last):
File "/home/qblocks/shashank/Development/Oct_21/Watermark_removal/deep-blind-watermark-removal_patch/main.py", line 71, in
main(args)
File "/home/qblocks/shashank/Development/Oct_21/Watermark_removal/deep-blind-watermark-removal_patch/main.py", line 41, in main
Machine.train(epoch)
File "/home/qblocks/shashank/Development/Oct_21/Watermark_removal/deep-blind-watermark-removal_patch/scripts/machines/VX.py", line 129, in train
l2_loss,att_loss,wm_loss,style_loss,ssim_loss = self.loss(outputs[0],self.norm(target),outputs[1],mask,outputs[2],self.norm(wm))
File "/home/qblocks/shashank/Development/Oct_21/Watermark_removal/wm/lib/python3.8/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
result = self.forward(*input, **kwargs)
File "/home/qblocks/shashank/Development/Oct_21/Watermark_removal/deep-blind-watermark-removal_patch/scripts/machines/VX.py", line 85, in forward
att_loss = self.attLoss(pred_ms, mask)
File "/home/qblocks/shashank/Development/Oct_21/Watermark_removal/wm/lib/python3.8/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
result = self.forward(*input, **kwargs)
File "/home/qblocks/shashank/Development/Oct_21/Watermark_removal/wm/lib/python3.8/site-packages/torch/nn/modules/loss.py", line 530, in forward
return F.binary_cross_entropy(input, target, weight=self.weight, reduction=self.reduction)
File "/home/qblocks/shashank/Development/Oct_21/Watermark_removal/wm/lib/python3.8/site-packages/torch/nn/functional.py", line 2525, in binary_cross_entropy
return torch._C._nn.binary_cross_entropy(
RuntimeError: CUDA error: device-side assert triggered

Initially I thought that this might be due to errors in the input mask as it expects the values to be b/w 0 and 1. However, upon printing the values of Mask and pred_ms (prediction), it was found that the model prediction tensor was NaN.

mask tensor([[[[1., 1., 1.,  ..., 0., 0., 0.],
          [1., 1., 0.,  ..., 0., 0., 0.],
          [0., 0., 0.,  ..., 0., 0., 0.],
          ...,
          [1., 1., 1.,  ..., 0., 0., 0.],
          [1., 1., 1.,  ..., 0., 0., 0.],
          [1., 1., 1.,  ..., 0., 0., 0.]]],


        [[[0., 0., 0.,  ..., 0., 0., 0.],
          [0., 0., 0.,  ..., 0., 0., 0.],
          [0., 0., 0.,  ..., 0., 0., 0.],
          ...,
          [0., 0., 0.,  ..., 0., 0., 0.],
          [0., 0., 0.,  ..., 0., 0., 0.],
          [0., 0., 0.,  ..., 0., 0., 0.]]],


        [[[0., 0., 0.,  ..., 0., 0., 0.],
          [0., 0., 0.,  ..., 0., 0., 0.],
          [0., 0., 0.,  ..., 0., 0., 0.],
          ...,
          [0., 0., 0.,  ..., 0., 0., 0.],
          [0., 0., 0.,  ..., 0., 0., 0.],
          [0., 0., 0.,  ..., 0., 0., 0.]]],


        [[[0., 0., 0.,  ..., 0., 0., 0.],
          [0., 0., 0.,  ..., 0., 0., 0.],
          [0., 0., 0.,  ..., 0., 0., 0.],
          ...,
          [1., 0., 0.,  ..., 1., 1., 1.],
          [1., 1., 0.,  ..., 1., 1., 0.],
          [1., 1., 1.,  ..., 1., 0., 0.]]]], device='cuda:0')

pred_mask tensor([[[[nan, nan, nan,  ..., nan, nan, nan],
          [nan, nan, nan,  ..., nan, nan, nan],
          [nan, nan, nan,  ..., nan, nan, nan],
          ...,
          [nan, nan, nan,  ..., nan, nan, nan],
          [nan, nan, nan,  ..., nan, nan, nan],
          [nan, nan, nan,  ..., nan, nan, nan]]],


        [[[nan, nan, nan,  ..., nan, nan, nan],
          [nan, nan, nan,  ..., nan, nan, nan],
          [nan, nan, nan,  ..., nan, nan, nan],
          ...,
          [nan, nan, nan,  ..., nan, nan, nan],
          [nan, nan, nan,  ..., nan, nan, nan],
          [nan, nan, nan,  ..., nan, nan, nan]]],


        [[[nan, nan, nan,  ..., nan, nan, nan],
          [nan, nan, nan,  ..., nan, nan, nan],
          [nan, nan, nan,  ..., nan, nan, nan],
          ...,
          [nan, nan, nan,  ..., nan, nan, nan],
          [nan, nan, nan,  ..., nan, nan, nan],
          [nan, nan, nan,  ..., nan, nan, nan]]],


        [[[nan, nan, nan,  ..., nan, nan, nan],
          [nan, nan, nan,  ..., nan, nan, nan],
          [nan, nan, nan,  ..., nan, nan, nan],
          ...,
          [nan, nan, nan,  ..., nan, nan, nan],
          [nan, nan, nan,  ..., nan, nan, nan],
          [nan, nan, nan,  ..., nan, nan, nan]]]], device='cuda:0',
       grad_fn=<SigmoidBackward>)

Following is the code that I am using to Crop the patch:

from __future__ import print_function, absolute_import

import os
import csv
import numpy as np
import json
import random
import math
import matplotlib.pyplot as plt
from collections import namedtuple
from os import listdir
from os.path import isfile, join

import torch
# torch.manual_seed(17)
import torch.utils.data as data

from scripts.utils.osutils import *
from scripts.utils.imutils import *
from scripts.utils.transforms import *
import torchvision.transforms as transforms
from PIL import Image
from PIL import ImageEnhance
from PIL import ImageFilter
from PIL import ImageFile
ImageFile.LOAD_TRUNCATED_IMAGES = True

import glob

class COCO(data.Dataset):
    def __init__(self,train,config=None, sample=[],gan_norm=False):

        self.train = []
        self.anno = []
        self.mask = []
        self.wm = []
        self.input_size = config.input_size
        self.normalized_input = config.normalized_input
        self.base_folder = config.base_dir
        self.dataset = train+config.data

        if config == None:
            self.data_augumentation = False
        else:
            self.data_augumentation = config.data_augumentation

        self.istrain = False if self.dataset.find('train') == -1 else True
        self.sample = sample
        self.gan_norm = gan_norm

        file_paths2 = sorted(glob.glob(join(self.base_folder,'wm_DIV2K','full_*',self.dataset,'image/*')))

        for fl2 in file_paths2:  
            file_name2 = fl2.split('/')[-1]
            self.train.append(fl2)
            self.mask.append(fl2.replace('/image/','/mask/'))
            self.wm.append(fl2.replace('/image/','/wm/'))
            self.anno.append(os.path.join(self.base_folder,'wm_DIV2K','natural',self.dataset,file_name2.split('-')[0]+'.'+file_name2.split('.')[-1]))

        if len(self.sample) > 0:
            self.train = [ self.train[i] for i in self.sample ] 
            self.mask = [ self.mask[i] for i in self.sample ] 
            self.anno = [ self.anno[i] for i in self.sample ] 

        self.trans = transforms.Compose([
                transforms.ToTensor(),
            ])

        print('total Dataset of '+self.dataset+' is : ', len(self.train))


    def __getitem__(self, index):
        img = Image.open(self.train[index]).convert('RGB')
        mask = Image.open(self.mask[index]).convert('L')
        anno = Image.open(self.anno[index]).convert('RGB')
        wm = Image.open(self.wm[index]).convert('RGB')

        W, H = img.size

        if W < self.input_size or H < self.input_size:
            img = img.resize((self.input_size, self.input_size))
            mask = mask.resize((self.input_size, self.input_size))
            anno = anno.resize((self.input_size, self.input_size))
            wm = wm.resize((self.input_size, self.input_size))
            

        i, j, h, w = transforms.RandomCrop.get_params(img, output_size=(self.input_size, self.input_size))
        img = transforms.functional.crop(img,i,j,h,w)
        mask = transforms.functional.crop(mask,i,j,h,w)
        anno = transforms.functional.crop(anno,i,j,h,w)
        wm = transforms.functional.crop(wm,i,j,h,w)
        
        
        img = self.trans(img)
        anno = self.trans(anno)
        mask = self.trans(mask)        
        wm = self.trans(wm)

        return {"image": img,
                "target": anno, 
                "mask": mask, 
                "wm": wm,
                "name": self.train[index].split('/')[-1],
                "imgurl":self.train[index],
                "maskurl":self.mask[index],
                "targeturl":self.anno[index],
                "wmurl":self.wm[index]
                }

    def __len__(self):

        return len(self.train)

Any Help would be appreciated!!

About Datasets/27kpng.zip

Hi, I did not find mask and wm folders after unzip the Datasets/27kpng.zip. Could you help me?

About Datasets/27kpng.zip

Hi, I did not find mask and wm folders after unzip the Datasets/27kpng.zip. Could you help me?

Try on my image

Hi, can I try it on watermarked image which I do not have mask. If so could you provide example on colab file removing watermark from uploaded image or from a test folder?

The results before and after running are different

code as follow:

from PIL import Image
img=Image.open('XX.jpg').resize((256,256))
image=np.array(img)
ih=torch.tensor(np.transpose(image,(2,0,1)))/255.0
ih=ih.unsqueeze(0)
imoutput,immask,imwatermark= model(ih.cuda())
imcoarser,imrefine,imwatermark = imoutput[1]immask + im(1-immask),imoutput[0]immask + im(1-immask),imwatermark*immask
torchvision.utils.save_image(imcoarser,'imrefine.jpg')

ModuleNotFoundError: No module named 'scripts.models.vmu2'

ModuleNotFoundError: No module named 'scripts.models.vmu2',
and how to inference one nature image ?

Other Pre-trained Models

Dear XiaoDong,

When will you be able to upload the pre-trained model on the grey image？

Look forward to your favourable reply.

27kpng_model_best.pth.tar The Compressed data has been corrupted

Hi, thank you for sharing your work.
I want to run your network on my data, But when I decompressed 27kpng_model_best.pth.tar, it says the file is corrupted, could you please upload it again？

About the forward func

1

Some question about dataset

Thanks for your impressive work.
I have some question about dataset in the onedrive link.
I guess:
10kmid.zip is LOGO-L
10khigh.zip is LOGO-H
10kgray.zip is LOGO-Gray
27kpng.zip is LOGO-30k
I wonder if this mapping is correct because there is some difference in naming.
Looking forward to your reply.

I can't find COCO_val2014.png

i tried your evaluate.sh. and downloaded your datasets, 10kgray, 10kmid
input "bash examples/evaluate.sh"
but got "FileNotFoundError: [Errno 2] No such file or directory: '/home/dell/watermark/10kgray/masks/natural/COCO_val2014.png'
What's "COCO_val2014.png"?

Supplementary Material

Kindly share your supplementary paper which has the model architecture changes shared mentioned in your paper. Thanks

Model unable to run on multiple GPUs

I have multiple GPUs and I would Like to train the model on them for faster training.
I see that you have already implemented MultiGPU training by using nn.DataParallel. There were some bugs in VX.py which were solved after i converted "self.model" to "self.model.module".

Yet after ensuring that i am using "CUDA_VISIBLE_DEVICES=0,1" i still see only GPU0 's memory to be filled and not GPU1's.

The model gives cuda out of memory if i try to use a input size >=512 with a batchsize of 12 or even 8.

Any idea why is it only consuming 1 of the 2 GPUs??

Thanks

how to train the model with my own data?

how to train the model with my own data?
Do you have the training script?

Thx

on video ?

would this model be available on videos ?

Support for variable sized input images?

Hi Vinthony,

I saw that you are resizing the input image to 256x256. Is there any support for variable size inputs?