Comments (7)
Sorry that I just see your comments yet.
And yes, we compute mean/std on training set, specifically, we first crop the original depth maps according to center points, which give us a hand-centered sub-image, then we compute the mean/std of all of these images.
from a2j.
def findMean(center):
trainingImageDir = './train_nyu/'
mean_D = 0
for i in tqdm(range(72757)): #72757
D_ = scio.loadmat(trainingImageDir + str(i+1) + '.mat')['depth']
D_ = np.array(D_,dtype=np.float)
new_Xmin = center[i][0][0] - cropHandWidth//2
new_Ymin = center[i][0][1] - cropHandHeight//2
new_Xmax = center[i][0][0] + cropHandWidth//2
new_Ymax = center[i][0][1] + cropHandHeight//2
D = D_[int(new_Ymin):int(new_Ymax), int(new_Xmin):int(new_Xmax)]
D[np.where(D >= center[i][0][2] + 200)] = center[i][0][2]
D[np.where(D <= center[i][0][2] - 200)] = center[i][0][2]
D -= center[i][0][2]
mean_D_ = D.mean(0).mean(0)
mean_D += mean_D_
# print(i)
mean_D /= 72757
return mean_D
I hope this helps.
from a2j.
Yes it really helps and I have understood. Thanks for your kind explanation! I will close this issue
from a2j.
Hello @zhangboshen, I tried to reproduce this but don't get the same results as yours. I believe it is the cropHandWidth
and cropHandHeight
. What values do you use for the above?
@DMCNerooo does yours match mean = -0.66877532
and std= 28.32958208
from a2j.
Hello @canuck6ix , I just want to understand the computation of mean/std in the data folder, but don't re-compute it. I use the value provided by the author. I think you can try to utilize above code "findMean" to get mean and then std, maybe there is little difference with author's but doesn't matter.
By the way, both cropHandWidth and cropHandHeight are set to 176 in this paper and project. Good luck!
from a2j.
@DMCNerooo Thanks for confirming width and height. I did utilize the code above and they are a bit different from the pre-computed ones. Here are the two for comparison:
autor's: mean = -0.67
std = 28.33
mine: mean = -3.86
std = 21.13
from a2j.
@DMCNerooo Thanks for confirming width and height. I did utilize the code above and they are a bit different from the pre-computed ones. Here are the two for comparison:
autor's:mean = -0.67
std = 28.33
mine:mean = -3.86
std = 21.13
@canuck6ix , I am sorry I just see your discuss now. And you can use your mean/std
, I think you can also get a resonable result.
from a2j.
Related Issues (20)
- How to obtain the center point coordinates and depth values during NYU inference HOT 4
- an you provide the mat file of the detection bounding boxes of the itop side and top training set
- Can you provide the mat file of the detection bounding boxes of the itop side and top training set HOT 2
- Problem while retraining A2J on NYU HOT 3
- Please add a requirements.txt file HOT 1
- 好奇下载的网络有没finetuning过 HOT 2
- Swapping Width and Height dimensions HOT 1
- Selecting anchor points with P = 0.02 HOT 2
- what's the mean of "depthFactor"? HOT 4
- Input Files for the ITOP dataset HOT 1
- Hands2017的数据是基于绝对3D坐标的,你们训练是基于UVD的, 请问Hands2017是如何训练的呢? HOT 6
- 训练ITOP_side数据集 HOT 1
- Unable to reproduce the results for the ITOP side view human body dataset. HOT 1
- reason for num_channel expansion? HOT 3
- training with missing keypoints? HOT 3
- How can I convert pre-trained model to coreml to use it on ios application?
- Why are the Anchor-Points generated on a grid with spaces between them? HOT 1
- Thank you so much!
- Hi, mat files are generated using this script: https://github.com/zhangboshen/A2J/blob/master/data/icvl/data_preprosess.m HOT 1
- Enquiry about drawing human 3D pose on our one depth image HOT 1
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 a2j.