File size: 2,707 Bytes
2b33fac
 
883f4ca
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2b33fac
6cc1858
 
 
2b33fac
6cc1858
 
2b33fac
6cc1858
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2b33fac
6cc1858
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2b33fac
6cc1858
2b33fac
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
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
---
library_name: keras
widget:
- text: input
  output:
    url: ./assets/input.png
- text: target
  output:
    url: ./assets/target.png
- text: output
  output:
    url: ./assets/output.png
metrics:
- TopIQ-FR
- ArcFace Cosine Distance
- VGGFace2 Cosine Distance
datasets:
- logasja/FDF
pipeline_tag: image-to-image
tags:
- adversarial
- aesthetic
- quality
- filter
base_model:
- vnet
- logasja/ArcFace
- logasja/VGGFace
license: gpl-3.0
---
<Gallery />

Training logs [here](https://wandb.ai/spuds/auramask/runs/62278206f0802ff9dc7648b4dee9ccfd)

# Model Description
This model uses a modified vnet for 2D input/output implemented [here](https://github.com/logasja/keras3-unets) with the following configuration.

```json
{
    "activation": "ReLU",
    "batch_norm": false,
    "filter_num": [
        128,
        256,
        512,
        1024,
        1024
    ],
    "n_labels": 3,
    "output_activation": "tanh",
    "pool": false,
    "res_num_ini": 1,
    "res_num_max": 3,
    "unpool": false
}
```

```json
{
    "alpha": 0.0001,
    "batch": 32,
    "epochs": 500,
    "epsilon": 1,
    "input": "(256, 256)",
    "losses": {
        "FEAT_ArcFace": {
            "d": "cosine_similarity",
            "f": "ArcFace",
            "name": "FEAT_ArcFace",
            "reduction": "sum_over_batch_size",
            "threshold": 0.68,
            "weight": 0.05
        },
        "FEAT_VGG-Face": {
            "d": "cosine_similarity",
            "f": "VGG-Face",
            "name": "FEAT_VGG-Face",
            "reduction": "sum_over_batch_size",
            "threshold": 0.68,
            "weight": 0.05
        },
        "IQASSIMC": {
            "lower_better": false,
            "name": "IQASSIMC",
            "reduction": "sum_over_batch_size",
            "weight": 0.5
        },
        "TopIQ": {
            "full_ref": true,
            "lower_better": false,
            "name": "TopIQ",
            "reduction": "sum_over_batch_size",
            "score_range": "~0, ~1",
            "weight": 0.5
        }
    },
    "mixed_precision": true,
    "optimizer": {
        "amsgrad": false,
        "beta_1": 0.9,
        "beta_2": 0.999,
        "clipnorm": null,
        "clipvalue": null,
        "ema_momentum": 0.99,
        "ema_overwrite_frequency": null,
        "epsilon": 1e-07,
        "global_clipnorm": null,
        "gradient_accumulation_steps": null,
        "learning_rate": 9.999999747378752e-05,
        "loss_scale_factor": null,
        "name": "adamw",
        "use_ema": false,
        "weight_decay": 0.004
    },
    "seed": "BIIIIIGSTRETCH",
    "testing": 0.01,
    "training": 0.99
}
```

## Model Architecture Plot
![](./assets/summary_plot.png)