christy1206 / ststnet Goto Github PK
View Code? Open in Web Editor NEWShallow Triple Stream Three-dimensional CNN
Shallow Triple Stream Three-dimensional CNN
Hi , could you please the link of the data you used. Other link is broken. I cannot wait to run and see the results. Thank you.
为什么我用给出的光流代码得到的光流图再进行实验,得到的UF1和UAR只有0.68左右呢?SAMM数据集我经过了裁剪,得到的光流特征没有已知数据集的光流特征更明显,请问是经过了什么处理吗?
May I ask why Python main.py -- train False appears after running it
X_ Train=torch Tensor (X_train). permute (0,3,1,2)
RuntimeError: number of dims don't match in permute
This is the result of the operation:
D:\STSTNet-master> python main.py --train False
lr=0.000050, epochs=800, device=cpu
Subject: 006
Traceback (most recent call last):
File "main.py", line 233, in
main(config)
File "main.py", line 153, in main
X_train = torch.Tensor(X_train).permute(0,3,1,2)
RuntimeError: number of dims don't match in permute
Dear author, it is my great honor to star your project. I would like to ask you how to use your project to recognize micro-expressions in videos or pictures
hi,where can I get the code of D&C-RoIs?
在进行LOSO交叉验证时,由于测试集比较小,得到的结果会不会存在偶然性呢?
When I use Google GPU to run Python code, I encounter this problem. How can I solve it? Hope to get your reply
lr=0.000050, epochs=800, device=cuda
Subject: sub25
Predicted : [0, 0, 0, 2, 2]
Ground Truth : [0, 0, 0, 2, 2]
Evaluation until this subject:
UF1: 1.0 | UAR: 1.0
Subject: sub26
Predicted : [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
Ground Truth : [1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0]
Evaluation until this subject:
UF1: 0.641 | UAR: 0.6667
Subject: sub20
Predicted : [0, 0]
Ground Truth : [0, 0]
Evaluation until this subject:
UF1: 0.6444 | UAR: 0.6667
Subject: sub21
Predicted : [1]
Ground Truth : [0]
Evaluation until this subject:
UF1: 0.6344 | UAR: 0.6444
Subject: sub24
Predicted : [2, 0, 0]
Ground Truth : [2, 0, 0]
Evaluation until this subject:
UF1: 0.6381 | UAR: 0.6471
Subject: sub22
Predicted : [0, 0]
Ground Truth : [0, 0]
Evaluation until this subject:
UF1: 0.641 | UAR: 0.6491
Subject: sub16
Predicted : [0, 1, 0]
Ground Truth : [1, 1, 0]
Evaluation until this subject:
UF1: 0.746 | UAR: 0.7333
Subject: sub19
Predicted : [2, 2, 2, 2, 2, 0, 0, 0, 1, 1, 1]
Ground Truth : [2, 2, 2, 2, 2, 0, 0, 0, 1, 1, 1]
Evaluation until this subject:
UF1: 0.8611 | UAR: 0.8427
Subject: sub17
Predicted : [2, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 1, 1, 1]
Ground Truth : [2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1]
Evaluation until this subject:
UF1: 0.9018 | UAR: 0.8903
Subject: sub23
Predicted : [0, 0, 0, 0, 0, 1, 0, 0]
Ground Truth : [0, 0, 0, 0, 0, 0, 0, 1]
Evaluation until this subject:
UF1: 0.8801 | UAR: 0.87
Subject: sub15
Predicted : [0, 2, 0]
Ground Truth : [1, 2, 0]
Evaluation until this subject:
UF1: 0.8695 | UAR: 0.8565
Subject: sub14
Predicted : [0, 1, 1]
Ground Truth : [1, 1, 1]
Evaluation until this subject:
UF1: 0.8722 | UAR: 0.8587
Subject: sub12
Predicted : [0, 0, 1, 0, 0, 1, 1, 2, 2, 2, 2]
Ground Truth : [0, 0, 0, 0, 0, 1, 1, 2, 2, 2, 2]
Evaluation until this subject:
UF1: 0.8757 | UAR: 0.8663
Subject: sub08
Predicted : [1]
Ground Truth : [0]
Evaluation until this subject:
UF1: 0.8672 | UAR: 0.8611
Subject: sub13
Predicted : [0, 1]
Ground Truth : [1, 1]
Evaluation until this subject:
UF1: 0.864 | UAR: 0.8563
Subject: sub09
Predicted : [0, 0, 0, 0, 0, 1, 0, 1, 1, 1]
Ground Truth : [0, 0, 0, 0, 0, 1, 1, 1, 1, 1]
Evaluation until this subject:
UF1: 0.8754 | UAR: 0.8672
Subject: sub07
Predicted : [1, 0, 0, 0, 0]
Ground Truth : [0, 0, 0, 0, 0]
Evaluation until this subject:
UF1: 0.8707 | UAR: 0.8643
Subject: sub06
Predicted : [2, 2, 0, 0]
Ground Truth : [2, 2, 1, 0]
Evaluation until this subject:
UF1: 0.8647 | UAR: 0.8569
Subject: sub11
Predicted : [0, 0, 0, 0]
Ground Truth : [0, 0, 0, 0]
Evaluation until this subject:
UF1: 0.8666 | UAR: 0.8584
Subject: sub05
Predicted : [2, 2, 0, 2, 2, 1]
Ground Truth : [2, 2, 2, 2, 2, 1]
Evaluation until this subject:
UF1: 0.8601 | UAR: 0.8463
Subject: sub04
Predicted : [0, 1]
Ground Truth : [0, 0]
Evaluation until this subject:
UF1: 0.8545 | UAR: 0.8427
Subject: sub03
Predicted : [0, 0, 0, 0, 2]
Ground Truth : [0, 0, 0, 0, 2]
Evaluation until this subject:
UF1: 0.8567 | UAR: 0.8449
Subject: s20
Traceback (most recent call last):
File "main.py", line 232, in
main(config)
File "main.py", line 152, in main
X_train = torch.Tensor(X_train).permute(0,3,1,2)
RuntimeError: number of dims don't match in permute
Hi!
I tried to repeat data preprocessing according to one in your paper. I've failed to get equal results.
I worked with CASME2 dataset and took cropped faces from CASME2_preprocessed_small_Li Xiaobai/Cropped.
I used the following code mainly from https://github.com/genbing99/SoftNet-SpotME repository:
import numpy as np
import pandas as pd
import cv2
def pol2cart(rho, phi): #Convert polar coordinates to cartesian coordinates for computation of optical strain
x = rho * np.cos(phi)
y = rho * np.sin(phi)
return (x, y)
def computeStrain(u, v):
u_x= u - pd.DataFrame(u).shift(-1, axis=1)
v_y= v - pd.DataFrame(v).shift(-1, axis=0)
u_y= u - pd.DataFrame(u).shift(-1, axis=0)
v_x= v - pd.DataFrame(v).shift(-1, axis=1)
os = np.array(np.sqrt(u_x**2 + v_y**2 + 1/2 * (u_y+v_x)**2).ffill(1).ffill(0))
return os
def resize(img, target_size=(28,28)):
rimg = cv2.resize(img, (*target_size,), interpolation=cv2.INTER_LINEAR)
return rimg
def perChannelNormalize(img):
res = img.copy()
for chnl in range(img.shape[2]):
channel = res[:,:,chnl]
res[:,:,chnl] = 255.*(channel - channel.min())/(channel.max() - channel.min())
return res.astype(np.uint8)
def get_optical_flow(img1, img2):
#Compute Optical Flow Features
assert img1.shape == img2.shape
# optical_flow = cv2.DualTVL1OpticalFlow_create() #Depends on cv2 version
optical_flow = cv2.optflow.DualTVL1OpticalFlow_create(
tau = 0.25,
#lambda_ = 0.15,
theta = 0.3,
nscales = 5,
warps=5,
epsilon=0.01,
innnerIterations=30,
outerIterations=10,
scaleStep=0.5,
gamma=0.1,
medianFiltering=5,
useInitialFlow=False
)
flow = optical_flow.calc(img1, img2, None)
magnitude, angle = cv2.cartToPolar(flow[..., 0], flow[..., 1])
u, v = pol2cart(magnitude, angle)
os = computeStrain(u, v)
#Features Concatenation
final = np.zeros((*img1.shape[0:2],3))
final[:,:,0] = os #B
final[:,:,1] = v #G
final[:,:,2] = u #R
return final
if __name__ == "__main__":
path1 = "img1.jpg"
path2 = "img2.jpg"
img1 = cv2.imread(path1, 0)
img2 = cv2.imread(path2, 0)
opfl = get_optical_flow(img1, img2) # get (u,v,os)
opfl = resize(opfl) # resize to 28x28
opfl = perChannelNormalize(opfl) # normalize each channel to 0..255
Help me please to get correct results!
Hi, I am wondering how to preprocess the raw data and transform them into norm_u_v_os.
Can you give me some advice or the preprocess code to this?
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.