Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,99 @@
|
|
1 |
-
---
|
2 |
-
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
language: en
|
3 |
+
tags:
|
4 |
+
- head-pose
|
5 |
+
- facial-landmarks
|
6 |
+
- computer-vision
|
7 |
+
license: mit
|
8 |
+
library_name: keras
|
9 |
+
model-index:
|
10 |
+
- name: Janus
|
11 |
+
results:
|
12 |
+
- task:
|
13 |
+
type: image-classification
|
14 |
+
name: Head Pose Recognition (Up/Down)
|
15 |
+
dataset:
|
16 |
+
name: Custom Dataset
|
17 |
+
type: custom
|
18 |
+
metrics:
|
19 |
+
- name: acc
|
20 |
+
type: accuracy
|
21 |
+
value: 0.904
|
22 |
+
- name: val_acc
|
23 |
+
type: accuracy
|
24 |
+
value: 0.9222
|
25 |
+
- task:
|
26 |
+
type: image-classification
|
27 |
+
name: Head Pose Recognition (Tilt)
|
28 |
+
dataset:
|
29 |
+
name: Custom Dataset
|
30 |
+
type: custom
|
31 |
+
metrics:
|
32 |
+
- name: acc
|
33 |
+
type: accuracy
|
34 |
+
value: 0.7655
|
35 |
+
- name: val_acc
|
36 |
+
type: accuracy
|
37 |
+
value: 0.9135
|
38 |
+
- task:
|
39 |
+
type: image-classification
|
40 |
+
name: Head Pose Recognition (Facing)
|
41 |
+
dataset:
|
42 |
+
name: Custom Dataset
|
43 |
+
type: custom
|
44 |
+
metrics:
|
45 |
+
- name: acc
|
46 |
+
type: accuracy
|
47 |
+
value: 0.9243
|
48 |
+
- name: val_acc
|
49 |
+
type: accuracy
|
50 |
+
value: 0.9462
|
51 |
+
pipeline_tag: image-classification
|
52 |
+
---
|
53 |
+
|
54 |
+
|
55 |
+
# Janus: Multi-Task Head Pose Recognition Model
|
56 |
+
|
57 |
+
## Model Description
|
58 |
+
|
59 |
+
Janus is a multi-class head pose recognition model that identifies key directional movements of the head. Named after the Roman god with two faces, Janus symbolizes different directions and perspectives. This model is designed for real-time applications such as gesture recognition, virtual meetings, and interactive systems.
|
60 |
+
|
61 |
+
### Key Features
|
62 |
+
|
63 |
+
- **Framework**: Keras
|
64 |
+
- **Model Type**: Functional (with combined multi-task outputs)
|
65 |
+
- **Input**: 468 facial landmarks (x, y, z coordinates) from MediaPipe
|
66 |
+
- **Output**: Classifications for head direction, motion, and up/down movement
|
67 |
+
- **Tasks**:
|
68 |
+
- Head Direction: Front, Left, Right
|
69 |
+
- Head Motion: Tilt, Turn
|
70 |
+
- Up/Down Motion: Up, Down, Unknown
|
71 |
+
|
72 |
+
## Limitations
|
73 |
+
|
74 |
+
- May not perform well on extreme or non-standard head poses
|
75 |
+
- Optimized primarily for front-facing camera angles
|
76 |
+
|
77 |
+
## Ethical Considerations
|
78 |
+
|
79 |
+
- Use responsibly and avoid applications that could misuse facial data
|
80 |
+
- Respect privacy and data protection regulations
|
81 |
+
|
82 |
+
## License
|
83 |
+
|
84 |
+
This model is released under the MIT License.
|
85 |
+
|
86 |
+
## Citation
|
87 |
+
|
88 |
+
If you use this model in your research, please cite:
|
89 |
+
|
90 |
+
```
|
91 |
+
@inproceedings{SUKO KUO_2024_janus,
|
92 |
+
title={Janus: Multi-Task Head Pose Recognition Model},
|
93 |
+
author={SUKO KUO},
|
94 |
+
year={2024},
|
95 |
+
url={https://huggingface.co/suko/Janus}
|
96 |
+
}
|
97 |
+
```
|
98 |
+
|
99 |
+
<!-- ChatGPT made me do it. I plead temporary AI-sanity. -->
|