File size: 5,931 Bytes
66388d5
 
1b6648d
 
 
 
 
 
 
 
 
 
 
 
66388d5
1b6648d
 
 
83f966b
 
c42dae7
 
8e1c908
1b6648d
a7099cc
1b6648d
 
 
 
5638dc5
 
1b6648d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
dc9b09f
1b6648d
dc9b09f
 
13cb42f
dc9b09f
29ad987
dc9b09f
 
 
 
 
42c21d5
dc9b09f
 
 
 
 
 
 
 
 
 
 
 
b4d7d2e
 
 
6d16450
6f26544
b4d7d2e
 
d1ec5b5
172f367
82f6e2b
 
 
3cf1b49
b183296
 
 
 
 
c991b6a
53c4640
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
---
license: mit
tags:
- thispersondoesnotexist
- stylegan
- stylegan2
- mesh
- model
- 3d
- asset
- generative
pretty_name: HeadsNet
size_categories:
- 1K<n<10K
---

# HeadsNet

The basic concept is to train a FNN/MLP on vertex data of 3D heads so that it can then re-produce random 3D heads.

This dataset uses the [thispersondoesnotexist_to_triposr_6748_3D_Heads](https://huggingface.co/datasets/tfnn/thispersondoesnotexist_to_triposr_6748_3D_Heads) dataset as a foundation.

The heads dataset was collecting using the scraper [Dataset_Scraper.7z](https://huggingface.co/datasets/tfnn/HeadsNet/resolve/main/Dataset_Scraper.7z?download=true) based on [TripoSR](https://github.com/VAST-AI-Research/TripoSR) which converts the 2D images from [ThisPersonDoesNotExist](https://thispersondoesnotexist.com/) into 3D meshes. _(using [this marching cubes improvement](https://github.com/VAST-AI-Research/TripoSR/issues/22#issuecomment-2010318709) by [thatname/zephyr](https://github.com/thatname))_

Vertex Normals need to be generated before we can work with this dataset, the easiest method to achieve this is with a simple [Blender](https://www.blender.org/) script:
```
import bpy
import glob
import pathlib
from os import mkdir
from os.path import isdir
importDir = "ply/"
outputDir = "ply_norm/"
if not isdir(outputDir): mkdir(outputDir)

for file in glob.glob(importDir + "*.ply"):
    model_name = pathlib.Path(file).stem
    if pathlib.Path(outputDir+model_name+'.ply').is_file() == True: continue
    bpy.ops.wm.ply_import(filepath=file)
    bpy.ops.wm.ply_export(
                            filepath=outputDir+model_name+'.ply',
                            filter_glob='*.ply',
                            check_existing=False,
                            ascii_format=False,
                            export_selected_objects=False,
                            apply_modifiers=True,
                            export_triangulated_mesh=True,
                            export_normals=True,
                            export_uv=False,
                            export_colors='SRGB',
                            global_scale=1.0,
                            forward_axis='Y',
                            up_axis='Z'
                        )
    bpy.ops.object.select_all(action='SELECT')
    bpy.ops.object.delete(use_global=False)
    bpy.ops.outliner.orphans_purge()
    bpy.ops.outliner.orphans_purge()
    bpy.ops.outliner.orphans_purge()
```
_Importing the PLY without normals causes Blender to automatically generate them._

At this point the PLY files now need to be converted to training data, for this I wrote a C program [DatasetGen_2_6.7z](https://huggingface.co/datasets/tfnn/HeadsNet/resolve/main/DatasetGen_2_6.7z?download=true) using [RPLY](https://w3.impa.br/~diego/software/rply/) to load the PLY files and convert them to binary data which I have provided here [HeadsNet-2-6.7z](https://huggingface.co/datasets/tfnn/HeadsNet/resolve/main/HeadsNet-2-6.7z?download=true).

It's always good to [NaN](https://en.wikipedia.org/wiki/NaN) check your training data after generating it so I have provided a simple Python script for that here [nan_check.py](https://huggingface.co/datasets/tfnn/HeadsNet/resolve/main/nan_check.py?download=true).

This binary training data can be loaded into Python using [Numpy](https://numpy.org/):
```
load_x = []
  with open("train_x.dat", 'rb') as f:
    load_x = np.fromfile(f, dtype=np.float32)

load_y = []
  with open("train_y.dat", 'rb') as f:
    load_y = np.fromfile(f, dtype=np.float32)
```

The data can then be reshaped and saved back out as a numpy array which makes for faster loading:
```
inputsize = 2
outputsize = 6
train_x = np.reshape(load_x, [tss, inputsize])
train_y = np.reshape(load_y, [tss, outputsize])
np.save("train_x.npy", train_x)
np.save("train_y.npy", train_y)
```

The basic premise of how this network is trained and thus how the dataset is generated in the C program is:
1. All models are pre-scaled to a normal cubic scale and then scaled again by 0.55 so that they all fit within a unit sphere.
2. All model vertices are reverse traced from the vertex position to the perimeter of the unit sphere using the vertex normal.
3. The nearest position on a 10,242 vertex icosphere is found and the network is trained to output the model vertex position and vertex color (6 components) at the index of the icosphere vertex.
4. The icosphere vertex index is scaled to a 0-1 range before being input to the network.
5. The network only has two input parameters, the other parameter is a 0-1 model ID which is randomly selected and all vertices for a specific model are trained into the network using the randomly selected ID. This ID does not change per-vertex it only changes per 3D model.
6. The ID allows the user to use this parameter as a sort of hyper-parameter for the random seed: to generate a random Head using this network you would input a random 0-1 seed and then iterate the icosphere index parameter to some sample range between 0-1 so if you wanted a 20,000 vertex head you would iterate between 0-1 at 20,000 increments of 0.00005 as the network outputs one vertex position and vertex color for each forward-pass.

* 1st input parameter = random seed
* 2nd input parameter = icosphere index

More about this type of network topology can be read here: https://gist.github.com/mrbid/1eacdd9d9239b2d324a3fa88591ff852

## Improvements
* Future networks will have 3 additional input parameters one for each x,y,z of a unit vector for the ray direction from the icosphere index.
* The unit vector used to train the network will just be the vertex normal from the 3D model but inverted.
* When performing inference more forward-passes would need to be performed as some density of rays in a 30° or similar cone angle pointing to 0,0,0 would need to be performed per icosphere index position.
* This could result in higher quality outputs.