Skip to content

HiNetPNGinPNGIccv2021

stegobox.codec.HiNetPNGinPNGIccv2021

Bases: BaseCodec

This steganography method is named: HiNetPNGinPNGIccv2021

  • Created by: Ruinan Ma
  • Created time: 2022/10/10

This is a PyTorch implementation of image steganography via deep learning, which is released in paper - HiNet: Deep Image Hiding by Invertible Network https://openaccess.thecvf.com/content/ICCV2021/html/Jing_HiNet_Deep_Image_Hiding_by_Invertible_Network_ICCV_2021_paper.html

Source code in stegobox/codec/hinet_pnginpng_iccv2021/hinet_pnginpng.py 
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
class HiNetPNGinPNGIccv2021(BaseCodec):
    """
    This steganography method is named: `HiNetPNGinPNGIccv2021`

    * Created by: Ruinan Ma
    * Created time: 2022/10/10

    This is a PyTorch implementation of image steganography via deep learning, which is
    released in paper - HiNet: Deep Image Hiding by Invertible Network
    https://openaccess.thecvf.com/content/ICCV2021/html/Jing_HiNet_Deep_Image_Hiding_by_Invertible_Network_ICCV_2021_paper.html
    """

    def __init__(
        self,
        para_pth: str = "ckpt/hinet_pnginpng_iccv2021/model.pt",
        verbose: bool = False,
    ) -> None:
        super().__init__()
        self.verbose = verbose
        self.para_pth = para_pth
        self.transform = transforms.Compose(
            [transforms.CenterCrop(c.CROPSIZE_VAL), transforms.ToTensor()]
        )
        self.dwt = common.DWT()
        self.iwt = common.IWT()
        self.HiNet = self.net_init()
        if verbose:
            if torch.cuda.is_available():
                print("Running HiNet_Iccv2021 with GPU.")
            else:
                print("Running HiNet_Iccv2021 with CPU.")

    def net_init(self):
        net = Model()
        net.cuda()
        init_model(net)
        net = torch.nn.DataParallel(net, device_ids=c.DEVICE_IDS)  # type: ignore
        params_trainable = list(filter(lambda p: p.requires_grad, net.parameters()))
        optim = torch.optim.Adam(
            params_trainable,
            lr=c.LR,
            betas=c.BETAS,
            eps=1e-6,
            weight_decay=c.WEIGHT_DECAY,
        )
        # weight_scheduler = torch.optim.lr_scheduler.StepLR(
        #     optim, c.weight_step, gamma=c.gamma
        # )
        load(self.para_pth, net, optim)
        net.eval()
        if self.verbose:
            print("HiNet loads pre_trained model successfully.")
        return net

    def encode(self, carrier: Image.Image, payload: Image.Image) -> torch.Tensor:
        """Encode image with format png into image with format png.

        Args:
            carrier: cover image
            payload: Payload secret image

        Returns:
            Encoded steganographic image with format torch.tensor
        """

        if self.verbose:
            print("Encoding...")
        carrier = self.transform(carrier)
        payload = self.transform(payload)

        # carrier.size()=payload.size()-->torch.Size([1, 3, 256, 256])
        carrier = carrier.unsqueeze(dim=0)  # type: ignore
        payload = payload.unsqueeze(dim=0)  # type: ignore

        # carrier_input.size()=payload_input.size()
        # -->torch.Size([1, 12, 128, 128])
        carrier_input = self.dwt(carrier)
        payload_input = self.dwt(payload)

        # input_img.size()-->torch.Size([1, 24, 128, 128])
        input_img = torch.cat([carrier_input, payload_input], dim=1)

        if torch.cuda.is_available():
            input_img = input_img.cuda()

        # output.size()-->torch.Size([1, 24, 128, 128])
        output = self.HiNet(input_img)

        # output_steg.size()-->torch.Size([1, 12, 128, 128])
        output_steg = output.narrow(1, 0, 4 * c.CHANNELS_IN)

        # output_z.size()-->torch.Size([1, 12, 128, 128])
        # output_z = output.narrow(
        #     1, 4 * c.channels_in, output.shape[1] - 4 * c.channels_in
        # )

        container = self.iwt(output_steg)
        # backward_z = gauss_noise(output_z.shape)
        return container

    def decode(self, carrier: Image.Image) -> torch.Tensor:
        """Decode secret image from encoded steganographic image.

        Args:
            carrier: Encoded carrier image.

        Returns:
            The decoded image if decode is successful.
        """

        if self.verbose:
            print("Decoding...")
        carrier = self.transform(carrier)
        carrier = carrier.unsqueeze(dim=0)  # type: ignore

        if torch.cuda.is_available():
            carrier = carrier.cuda()  # type: ignore

        carrier = self.dwt(carrier)
        backward_z = gauss_noise(carrier.shape)  # type: ignore
        output_rev = torch.cat([carrier, backward_z], dim=1)  # type: ignore
        bacward_img = self.HiNet(output_rev, rev=True)
        secret_rev = bacward_img.narrow(
            1, 4 * c.CHANNELS_IN, bacward_img.shape[1] - 4 * c.CHANNELS_IN
        )
        reveal_img = self.iwt(secret_rev)
        return reveal_img

encode(carrier, payload)

Encode image with format png into image with format png.

Parameters:

Name Type Description Default
carrier Image

cover image

 required
payload Image

Payload secret image

 required

Returns:

Type Description
Tensor

Encoded steganographic image with format torch.tensor
Source code in stegobox/codec/hinet_pnginpng_iccv2021/hinet_pnginpng.py 
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
def encode(self, carrier: Image.Image, payload: Image.Image) -> torch.Tensor:
    """Encode image with format png into image with format png.

    Args:
        carrier: cover image
        payload: Payload secret image

    Returns:
        Encoded steganographic image with format torch.tensor
    """

    if self.verbose:
        print("Encoding...")
    carrier = self.transform(carrier)
    payload = self.transform(payload)

    # carrier.size()=payload.size()-->torch.Size([1, 3, 256, 256])
    carrier = carrier.unsqueeze(dim=0)  # type: ignore
    payload = payload.unsqueeze(dim=0)  # type: ignore

    # carrier_input.size()=payload_input.size()
    # -->torch.Size([1, 12, 128, 128])
    carrier_input = self.dwt(carrier)
    payload_input = self.dwt(payload)

    # input_img.size()-->torch.Size([1, 24, 128, 128])
    input_img = torch.cat([carrier_input, payload_input], dim=1)

    if torch.cuda.is_available():
        input_img = input_img.cuda()

    # output.size()-->torch.Size([1, 24, 128, 128])
    output = self.HiNet(input_img)

    # output_steg.size()-->torch.Size([1, 12, 128, 128])
    output_steg = output.narrow(1, 0, 4 * c.CHANNELS_IN)

    # output_z.size()-->torch.Size([1, 12, 128, 128])
    # output_z = output.narrow(
    #     1, 4 * c.channels_in, output.shape[1] - 4 * c.channels_in
    # )

    container = self.iwt(output_steg)
    # backward_z = gauss_noise(output_z.shape)
    return container

decode(carrier)

Decode secret image from encoded steganographic image.

Parameters:

Name Type Description Default
carrier Image

Encoded carrier image.

 required

Returns:

Type Description
Tensor

The decoded image if decode is successful.
Source code in stegobox/codec/hinet_pnginpng_iccv2021/hinet_pnginpng.py 
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
def decode(self, carrier: Image.Image) -> torch.Tensor:
    """Decode secret image from encoded steganographic image.

    Args:
        carrier: Encoded carrier image.

    Returns:
        The decoded image if decode is successful.
    """

    if self.verbose:
        print("Decoding...")
    carrier = self.transform(carrier)
    carrier = carrier.unsqueeze(dim=0)  # type: ignore

    if torch.cuda.is_available():
        carrier = carrier.cuda()  # type: ignore

    carrier = self.dwt(carrier)
    backward_z = gauss_noise(carrier.shape)  # type: ignore
    output_rev = torch.cat([carrier, backward_z], dim=1)  # type: ignore
    bacward_img = self.HiNet(output_rev, rev=True)
    secret_rev = bacward_img.narrow(
        1, 4 * c.CHANNELS_IN, bacward_img.shape[1] - 4 * c.CHANNELS_IN
    )
    reveal_img = self.iwt(secret_rev)
    return reveal_img