Comments (11)
Hi, if you use depth cameras other than SR300, the only thing you have to deal with is changing the intrinsic parameters in hand_pose_estimator.cpp or realsense_realtime_demo_librealsense2.py.
The default values are intrinsic parameters of Intel SR300 camera and you can replace these parameters with those of R415 or 435.
Another tip for real-time demo: since we simply use a naive depth thresholding strategy to detect and segment the hand, you have to put your right hand in infront of the camera and try to avoid clusttered foreground and redundant arm around the hand.
Currently I am not quite sure if we could release the pre-trained models on Hands17 dataset online due to the related license of the dataset. We may contact the authors of the dataset to clarify this issue in the future.
from pose-ren.
@joohaeng Hi, In cases you are still interested, I just release the pre-trained models on HANDS17 dataset online.
from pose-ren.
@Suraj520 Would you mind checking the original depth image from the camera to see if the camera is working properly.
According to your screenshot, the depth image is quite poor and the hand can hardly be recognized even by eyes. In the display function, I manually discard the depth values (> 1500) that are too far away from the camera. Did you happen to put your hand too far away?
Also, if you set the parameters lower_ and upper_ to 180 and 350, you should make sure your hand is within [180mm, 350mm] from the camera, otherwise the hand segmentation would fail.
from pose-ren.
I could run the demo with SR300, which shows the great performance as in your demo images.
I will also test with the R400 cameras with proper intrinsic parameters. By the way, do you expect R400 will give some performance gain in pose estimation?
I've already have Hands17 dataset. Before waiting your release in the future, I may try to train with this dataset by myself. Do you have any guides for training procedure with Hands17 dataset?
Thank you again.
from pose-ren.
I am not quite familiar with R400 camera, but I think it will bring some performance gain if it can provide better depth images than SR300.
You can find details of the training procedure in our paper.
from pose-ren.
I could run the real-time demo for D435 with the following intrinsic values after calibration:
stream.depth: width: 640, height: 480, ppx: 316.802, ppy: 241.818, fx: 385.13, fy: 385.13
The performance was as good as SR300, I guess.
By the way, what should be proper values for lower_ and upper_? The default values are 0 and 650. But, I changed them to 180 and 350.
Thanks again.
from pose-ren.
Thanks for sharing the intrinsic parameters for D435 camera! Glad to hear that the demo works good for D435.
The values of lower_
and upper_
are just used for hand segmentation via depth thresholding. It's ok to change them to any other values as long as the hand is properly segmented.
from pose-ren.
@joohaeng Hi, In cases you are still interested, I just release the pre-trained models on HANDS17 dataset online.
That's HUGE! I was testing MSR models to recognize ASL. But found some limitations for subtle finger postures. I am so curious to know how HANDS17 performs for harder cases. Thanks a lot!
from pose-ren.
I could run the real-time demo for D435 with the following intrinsic values after calibration:
stream.depth: width: 640, height: 480, ppx: 316.802, ppy: 241.818, fx: 385.13, fy: 385.13The performance was as good as SR300, I guess.
By the way, what should be proper values for lower_ and upper_? The default values are 0 and 650. But, I changed them to 180 and 350.
Thanks again.
Hi @joohaeng , I am trying to run the demo of Pose-REN using Intel Realsense D435, I was redirected by @xinghaochen to use the camera intrinsic parameters mentioned by you. I have a small doubt - "Are the ux, uy mentioned in the Line 53 of "realsense_realtime_demo_librealsense2.py " same as the values of ppx and ppy mentioned by you above !?,
because I modified the scripts according to your suggestion above with the instrinsic parameters of D435 but the hand segmentation and pose estimation is not appreciable (*P.S: I am following all the rest instructions of using right hand, non occluded background etc mentioned by @xinghaochen .)
Looking forward for your help!.
(Attaching my detection results using icvl pre-trained models on Intel realsense D435)
from pose-ren.
Hi @joohaeng , I am trying to run the demo of Pose-REN using Intel Realsense D435, I was redirected by @xinghaochen to use the camera intrinsic parameters mentioned by you. I have a small doubt - "Are the ux, uy mentioned in the Line 53 of "realsense_realtime_demo_librealsense2.py " same as the values of ppx and ppy mentioned by you above !?,
because I modified the scripts according to your suggestion above with the instrinsic parameters of D435 but the hand segmentation and pose estimation is not appreciable (*P.S: I am following all the rest instructions of using right hand, non occluded background etc mentioned by @xinghaochen .)Looking forward for your help!.
(Attaching my detection results using icvl pre-trained models on Intel realsense D435)
Hi @Suraj520. My experiment with D435 was a month ago, and I've got a bad memory. ^_^
I'll check my environment and be back to you soon. Thanks.
from pose-ren.
@Suraj520 Would you mind checking the original depth image from the camera to see if the camera is working properly.
According to your screenshot, the depth image is quite poor and the hand can hardly be recognized even by eyes. In the display function, I manually discard the depth values (> 1500) that are too far away from the camera. Did you happen to put your hand too far away?
Also, if you set the parameters lower_ and upper_ to 180 and 350, you should make sure your hand is within [180mm, 350mm] from the camera, otherwise the hand segmentation would fail.
Thanks a lot @xinghaochen , your suggestions helped me to solve the bug! :)
from pose-ren.
Related Issues (20)
- What do you mean by those icvl_center.txt and msra_center.txt files? HOT 17
- About Data Layer HOT 4
- hi HOT 1
- caffe-pose HOT 1
- the result of kinect realtime test is very terrible! HOT 4
- build demo error HOT 4
- Hi. I have some question about data augmentation and total learning time. HOT 1
- Can I test both hands? HOT 1
- what does “annotation dir” mean? I can’t find this dir in caffe-pose
- caffe-pose HOT 1
- About format training data
- CUDA error while loading network HOT 1
- Training code HOT 2
- Running with hands17 net HOT 1
- about hand size and data augmentation HOT 5
- FPS and system specification HOT 1
- About training HOT 3
- src/testing/predict.py HOT 5
- Will the caffe be essentially first-built? HOT 3
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from pose-ren.