Facial_KeyPoint training using TLT

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

Hi,
I am curious to know if “Facial landmark training” is possible using TLT?

Thank you

3

Yes, see https://docs.nvidia.com/tao/tao-toolkit/text/facial_landmarks_estimation/index.html

4

Thank you!
I would like to know, how can I download the jupyter-notebook and other requirements?

5

See TAO Toolkit Quick Start Guide — TAO Toolkit 3.22.05 documentation

6

I mean which Model architecture name should I use, is it FACIAL LANDMARK which is FPENET or GAZENET?

7

Refer to Facial Landmarks Estimation — TAO Toolkit 3.21.11 documentation

fpenet.

8

Which tool is recommended for facial keypoint annotations?

Label Studio is giving out in the below format
[
{
“img”: “/data/upload/2/8140812d-IMG-20210930-WA0107_Ref.jpg”,
“id”: 11,
“kp-1”: [
{
“x”: 45.333333333333336,
“y”: 47.4,
“width”: 0.26666666666666666,
“keypointlabels”: [
“Nose”
],
“original_width”: 864,
“original_height”: 1152
},
{
“x”: 48.8,
“y”: 46.5,
“width”: 0.26666666666666666,
“keypointlabels”: [
“Nose”
],
“original_width”: 864,
“original_height”: 1152
},
{
“x”: 42.53333333333333,
“y”: 46.8,
“width”: 0.26666666666666666,
“keypointlabels”: [
“Nose”
],
“original_width”: 864,
“original_height”: 1152
},
{
“x”: 51.2,
“y”: 44.6,
“width”: 0.26666666666666666,
“keypointlabels”: [
“Nose”
],
“original_width”: 864,
“original_height”: 1152
},
{
“x”: 49.86666666666667,
“y”: 42.4,
“width”: 0.26666666666666666,
“keypointlabels”: [
“Nose”
],
“original_width”: 864,
“original_height”: 1152
},
{
“x”: 48.8,
“y”: 40.6,
“width”: 0.26666666666666666,
“keypointlabels”: [
“Nose”
],
“original_width”: 864,
“original_height”: 1152
},
{
“x”: 41.86666666666667,
“y”: 44.6,
“width”: 0.26666666666666666,
“keypointlabels”: [
“Nose”
],
“original_width”: 864,
“original_height”: 1152
},
{
“x”: 43.86666666666667,
“y”: 42.6,
“width”: 0.26666666666666666,
“keypointlabels”: [
“Nose”
],
“original_width”: 864,
“original_height”: 1152
},
{
“x”: 44.53333333333333,
“y”: 39.8,
“width”: 0.26666666666666666,
“keypointlabels”: [
“Nose”
],
“original_width”: 864,
“original_height”: 1152
}
],
“annotator”: 1,
“annotation_id”: 3,
“created_at”: “2021-12-14T09:59:25.569325Z”,
“updated_at”: “2021-12-14T09:59:34.222382Z”,
“lead_time”: 179.249
}
]

9

If Label Studio was used for annotating the facial points then how to convert it from json file of label studio into tlt’s FPENET’s format

10

Please refer to Facial Landmarks Estimation | NVIDIA NGC

The Sloth and Label-Studio tools have been utilized for labeling.

Please download jupyter notebook for reference.

or see GPU-optimized AI, Machine Learning, & HPC Software | NVIDIA NGC directly.

12

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.