Emiliana
|
a0ccd9a162
|
Reduced the impact of eye blinks and opening the mouth on head pose and added new models to benchmark.
|
2020-12-07 18:11:44 +01:00 |
|
Emiliana
|
2f7d8b04dc
|
Fixed a possible division by zero.
|
2020-11-28 15:54:06 +01:00 |
|
Emiliana
|
400e12f334
|
Integrated two new tracking models.
|
2020-11-28 15:50:22 +01:00 |
|
Emiliana
|
8354e2829a
|
Changed eye features to use the absolute value rather than clipped ones.
|
2020-10-26 21:16:39 +01:00 |
|
Emiliana
|
3038cad92b
|
Modified how the mouth open feature is calculated.
|
2020-10-18 20:35:55 +02:00 |
|
Emiliana
|
e805aa24e3
|
Reverted "Switched back to decoding face landmarks within the models."
This reverts commit 4038d334b1.
|
2020-10-08 13:07:51 +02:00 |
|
Emiliana
|
4e3ee5795d
|
Added a new --try-hard flag.
|
2020-10-06 23:44:23 +02:00 |
|
Emiliana
|
4038d334b1
|
Switched back to decoding face landmarks within the models.
|
2020-10-05 19:57:57 +02:00 |
|
Emiliana
|
d9cd16b095
|
Fixed Blackmagic capture.
|
2020-09-30 19:31:53 +02:00 |
|
Emiliana
|
d8c8ed05e8
|
Added --benchmark option and disabled gaze tracking for models with negative IDs.
|
2020-09-26 23:49:54 +02:00 |
|
Emiliana
|
1aca1bee66
|
Fixed gaze tracking issues introduced recently.
|
2020-09-05 15:05:32 +02:00 |
|
Emiliana
|
0d61b061a3
|
Added a separate detection threshold parameter.
|
2020-09-02 18:42:25 +02:00 |
|
Emiliana
|
bde32b7213
|
Added support for 30 point tracking to OpenSeeShowPoints and OpenSeeVRMDriver.
|
2020-09-02 15:31:35 +02:00 |
|
Emiliana
|
fb00f6aff1
|
Added basic support for toaster level tracking model.
|
2020-09-01 18:53:15 +02:00 |
|
Emiliana
|
87236c0549
|
Fixed bug with --gaze-tracking 0 and added explicit gc calls at end of frame.
|
2020-08-21 22:47:02 +02:00 |
|
Emiliana
|
9ce7d39505
|
Added filters for NaN values.
|
2020-08-11 16:27:20 +02:00 |
|
Emiliana
|
c9eb3c1826
|
Removed landmark detection from ONNX model due to potential compatibility issues.
|
2020-08-11 10:44:07 +02:00 |
|
Emiliana
|
2fc0f52385
|
Added camera reinitialization.
|
2020-08-10 22:09:17 +02:00 |
|
Emiliana
|
8dbc573830
|
Reduced size of binary after scikit-image made it double.
|
2020-08-10 18:05:09 +02:00 |
|
Emiliana
|
878c6b8833
|
Reduced unnecessary iteration.
|
2020-08-10 15:42:18 +02:00 |
|
Emiliana
|
50224a1e2e
|
Slightly optimized eye alignment and landmark decoding.
|
2020-08-10 14:40:40 +02:00 |
|
Emiliana
|
adbf3a4bed
|
Merge branch 'master' of github.com:emilianavt/OpenSeeFace
# Conflicts:
# tracker.py
|
2020-08-10 00:13:01 +02:00 |
|
Emiliana
|
96ca687d0c
|
Moved landmark decoding into the ONNX model.
|
2020-08-10 00:03:24 +02:00 |
|
Emiliana
|
76bd3dcf3a
|
Optimized estimate_depth function for a 0.5ms speedup.
|
2020-08-09 19:34:14 +02:00 |
|
Emiliana
|
6b876be623
|
Optimized landmarks function for a 1-2ms speedup.
|
2020-08-09 17:32:29 +02:00 |
|
Emiliana
|
d8d0b0bff0
|
When --max-feature-updates is enabled, feature stats are no longer discarded upon detecting a new face.
|
2020-06-25 17:19:51 +02:00 |
|
Emiliana
|
0690bdd15e
|
Face 3D point adaption can be disabled and feature adaptability can be limited now.
|
2020-06-24 12:29:33 +02:00 |
|
Emiliana
|
ea735ffdde
|
This adds a new medium quality model. Unused models were removed and the README.md file updated.
|
2020-03-29 15:56:54 +02:00 |
|
Emiliana
|
8c547a3b39
|
Some slight optimizations for multi face tracking.
|
2020-03-28 14:57:56 +01:00 |
|
Emiliana
|
b6554ddeaa
|
This adds a new fastest model.
|
2020-03-28 13:14:50 +01:00 |
|
Emiliana
|
4268abb00a
|
The basic 3D point model used for 3D point fitting is now based on data from fitting the points to head-pose-face-detection-female.mp4 over multiple iterations
|
2020-03-26 19:00:33 +01:00 |
|
Emiliana
|
60fb3aac31
|
Lower overlap threshold.
|
2020-03-25 23:02:23 +01:00 |
|
Emiliana
|
9bd3fcc617
|
When multiple faces were close to each other, some of them were removed by accident. This fixes that.
|
2020-03-25 22:52:28 +01:00 |
|
Emiliana
|
5c118347b6
|
New and improved tracking model and confidence value visualization.
|
2020-03-25 17:03:30 +01:00 |
|
Emiliana
|
14b4bd9122
|
As the Haar Cascade face detector is no longer used, it is now remoed.
|
2020-03-22 14:14:26 +01:00 |
|
Emiliana
|
c03e222a49
|
Smaller, faster and better detection model.
|
2020-03-22 13:47:55 +01:00 |
|
Emiliana
|
a10eacc670
|
A new, custom face detection model is now used to lock onto faces for tracking.
|
2020-03-22 00:34:34 +01:00 |
|
Emiliana
|
31536929f7
|
These changes further improve point normalization to make it easier to detect expressions under rotations.
|
2020-03-08 18:05:12 +01:00 |
|
Emiliana
|
0c84f50672
|
3D points are now normalized by reference sizes.
|
2020-03-01 22:27:50 +01:00 |
|
Emiliana
|
04d7f5b8b3
|
This adds manually designed features to the OpenSeeData.
|
2020-03-01 14:23:56 +01:00 |
|
Emiliana
|
04e19c9798
|
This adds optimizations and safety checks to the 3D point adjustments and adds an option to the IK target, which allows its position to slowly drift back to neutral.
|
2020-02-28 20:19:03 +01:00 |
|
Emiliana
|
033a48b132
|
This changes the points used for PnP solving and replaces the RetinaFace model with a higher resolution version.
|
2020-02-28 02:10:23 +01:00 |
|
Emiliana
|
cd9cfd9ee8
|
RetinaFace was added as a face detection scheme.
|
2020-02-26 00:54:32 +01:00 |
|
Emiliana
|
44060dfff2
|
The eye tracking model was switched out. This one seems to be more stable.
|
2020-02-24 21:51:54 +01:00 |
|
Emiliana
|
0332f9cb93
|
This adds the gaze tracker.
|
2020-02-24 00:44:17 +01:00 |
|
Emiliana
|
e60b9b37bd
|
The internal face 3D model will now adapt to the tracked landmarks.
|
2020-02-20 21:33:00 +01:00 |
|
Emiliana
|
83e400a1d1
|
Face ids can now be displayed.
|
2020-02-08 00:13:28 +01:00 |
|
Emiliana
|
7998bf715c
|
The code is now refactored to use the new FaceInfo class instead of passing around big tuples. Keeping track of which face is which is more reliable now too.
|
2020-02-07 20:59:46 +01:00 |
|
Emiliana
|
0a1bedb352
|
The face tracker now keeps track of which face is which. This makes 3D point estimation for multi face scenarios both more stable and faster.
|
2020-02-07 00:55:29 +01:00 |
|
Emiliana
|
423b5c6c86
|
The model selection parameter should now be easier to understand and two less useful models have been removed to save space. Width, height and FPS are now written to log file and both width and height are also transmitted over the network. The video output function has been changed to use a lossless codec, but videos need to be converted to a more common format before importing them to Vegas now.
|
2020-02-07 00:09:48 +01:00 |
|