search

andrew-pengjj / Enhanced-3DTV

The code of enhanced 3DTV Regularization and Its Applications on Hyper-spectral Image Denoising and Compressed Sensing
55 stars 12 forks source link

Fail to reproduce the results of Washington DC Mall in Compress Sensing task #1

Open Zeqiang-Lai opened 3 years ago

Zeqiang-Lai commented 3 years ago

testdata: x_dc.mat

sampling ratio is 0.2

parameters are set by default:

Rk           = 7;
weight       = [0.015,0.015,0.015];

log of last two iteration

**********************iter: 96 ***********************
stopFormulaVal: 1.1146e-03 
stopFormulaVal2: 1.5997e+01 
beta value is: 0.309133, 0.309133, 0.408828, 4.088284
The clean image: mp = 25.112513, sm = 0.606298, er =1140.122523.
The full image: mp = 25.815700, sm = 0.556139, er =1076.321365.

**********************iter: 97 ***********************
stopFormulaVal: 0.0000e+00 
stopFormulaVal2: 1.5997e+01 
 !!!stopped by termination rule!!!
andrew-pengjj commented 1 year ago

Hi, Zeqiang-Lai.

This is because the given data x_dc data type is int16, and after trans255, it becomes a binary distribution (0, 255) image. Here you only need to change the 13th line of demo.m to x=trans255(double(x_dc)) to run normally.

Thank you for your reminder！

colorfuloranges commented 9 months ago

testdata: x_dc.mat

sampling ratio is 0.2

parameters are set by default:

Rk           = 7;
weight       = [0.015,0.015,0.015];

log of last two iteration

**********************iter: 96 ***********************
stopFormulaVal: 1.1146e-03 
stopFormulaVal2: 1.5997e+01 
beta value is: 0.309133, 0.309133, 0.408828, 4.088284
The clean image: mp = 25.112513, sm = 0.606298, er =1140.122523.
The full image: mp = 25.815700, sm = 0.556139, er =1076.321365.

**********************iter: 97 ***********************
stopFormulaVal: 0.0000e+00 
stopFormulaVal2: 1.5997e+01 
 !!!stopped by termination rule!!!

hello,can you share Washington DC Mall dataset with me? I found this data set on the Internet, but there are some problems with the ground truth map. Could you please share it with me? My email is forevergbin@163.com. thank you!!!

andrew-pengjj commented 9 months ago

I'm aware of your problem. The reason for this phenomenon may be related to the parameters you selected.

The parameter of the model is:

Rk =7 weight= [0.0075,0.0075,0.0075];

The result of 48 iteration is:

**iter: 47 *** stopFormulaVal: 3.4648e-03 stopFormulaVal2: 1.7965e+02 beta value is: 0.008438, 0.011159, 0.014758, 0.097035 The clean image: mp = 45.579994, sm = 0.995430, er =22.788762. The full image: mp = 46.002006, sm = 0.995583, er =21.745532.

**iter: 48 *** stopFormulaVal: 3.1959e-03 stopFormulaVal2: 1.6989e+02 beta value is: 0.009704, 0.012833, 0.016972, 0.097035 The clean image: mp = 45.745716, sm = 0.995523, er =22.502498. The full image: mp = 46.138524, sm = 0.995657, er =21.524360.

If you want to get a better-recovered result, you need to fine-tune the parameters. There is a parameter tuning tip: if the sampling rate is relatively high, you should choose a larger rank and smaller weight parameters.

colorfuloranges commented 9 months ago

I'm aware of your problem. The reason for this phenomenon may be related to the parameters you selected.

The parameter of the model is:

Rk =7 weight= [0.0075,0.0075,0.0075];

The result of 48 iteration is:

**iter: 47 *** stopFormulaVal: 3.4648e-03 stopFormulaVal2: 1.7965e+02 beta value is: 0.008438, 0.011159, 0.014758, 0.097035 The clean image: mp = 45.579994, sm = 0.995430, er =22.788762. The full image: mp = 46.002006, sm = 0.995583, er =21.745532.

**iter: 48 *** stopFormulaVal: 3.1959e-03 stopFormulaVal2: 1.6989e+02 beta value is: 0.009704, 0.012833, 0.016972, 0.097035 The clean image: mp = 45.745716, sm = 0.995523, er =22.502498. The full image: mp = 46.138524, sm = 0.995657, er =21.524360.

If you want to get a better-recovered result, you need to fine-tune the parameters. There is a parameter tuning tip: if the sampling rate is relatively high, you should choose a larger rank and smaller weight parameters.

您好 同学 我想要 Washington DC Mall dataset 这个数据集 我在网上下载的数据集没有真值图 我想要一下这个数据的真值图（地物分类的真实标签）请问您有吗？

andrew-pengjj commented 9 months ago

真值图我没有，我一般做分类，用的是indian pines和PaviaU这两个数据。indian pines的数据我有真值图。您需要可以添加我的微信：pjj_eospeng.

colorfuloranges @.***> 于2024年1月21日周日 19:24写道：

I'm aware of your problem. The reason for this phenomenon may be related to the parameters you selected.

The parameter of the model is:

Rk =7 weight= [0.0075,0.0075,0.0075];

The result of 48 iteration is:

**iter: 47 *** stopFormulaVal: 3.4648e-03 stopFormulaVal2: 1.7965e+02 beta value is: 0.008438, 0.011159, 0.014758, 0.097035 The clean image: mp = 45.579994, sm = 0.995430, er =22.788762. The full image: mp = 46.002006, sm = 0.995583, er =21.745532.

**iter: 48 *** stopFormulaVal: 3.1959e-03 stopFormulaVal2: 1.6989e+02 beta value is: 0.009704, 0.012833, 0.016972, 0.097035 The clean image: mp = 45.745716, sm = 0.995523, er =22.502498. The full image: mp = 46.138524, sm = 0.995657, er =21.524360.

If you want to get a better-recovered result, you need to fine-tune the parameters. There is a parameter tuning tip: if the sampling rate is relatively high, you should choose a larger rank and smaller weight parameters.

您好 同学 我想要 Washington DC Mall dataset 这个数据集 我在网上下载的数据集没有真值图 我想要一下这个数据的真值图（地物分类的真实标签）请问您有吗？

— Reply to this email directly, view it on GitHub https://github.com/andrew-pengjj/Enhanced-3DTV/issues/1#issuecomment-1902598540, or unsubscribe https://github.com/notifications/unsubscribe-auth/AGK3CUOTWBKI64R6EBVUKMDYPT3FNAVCNFSM4X34HVY2U5DIOJSWCZC7NNSXTN2JONZXKZKDN5WW2ZLOOQ5TCOJQGI2TSOBVGQYA . You are receiving this because you commented.Message ID: @.***>