Study performance with under-sampled and noisy data

[tobias-liaudat]

Perform a study with under-sampled stars (aka with real Euclid sampling) and with noisy data of different SNR values (each SNR taken randomly from a valid SNR range).

• [ ] Study the performance change if we use dithers (x4 stars).

• [x] Need to evaluate the e1, e2 and R2 RMSE and compare them to Euclid requirements.

[tobias-liaudat]

No-dithers and comparison on under-sampled images

Example of the different noise levels

Example of some stars

Quantitative result table

model	tr pix RMSE	te pix RMSE	tr OPD RMSE	te OPD RMSE
parametric_cycle1	1.1087e-03	1.0950e-03	7.5148e-02	7.6320e-02
-	-	-	-	-
poly_cycle1	3.3905e-04	4.0550e-04	1.1021e-01	1.0950e-01
poly_cycle2	2.7491e-04	3.1410e-04	1.1339e-01	1.1409e-01
-	-	-	-	-
graph_poly_cycle1	2.4031e-04	2.7927e-04	1.3375e-01	1.3413e-01
graph_poly_cycle2	2.9310e-04	3.2360e-04	1.2921e-01	1.3020e-01

[tobias-liaudat]

Shape and size results

We calculate the shape and size metric on polychromatic PSFs with a 100% knowledge of the object's SED. The sampling of the PSFs is of 3x times the Euclid natural resolution in order to be sure we are Nyquist sampled at all the wavelengths of the VIS instrument pass-band. The error is computed with respect to the ground truth PSF model. The shape measurement is done with the HSM module of the Galsim package. The algorithm is known as adaptive moments.

Euclid requirements are:

• sigma(e1) < 2x10^-4
• sigma(e2) < 2x10^-4
• sigma(R2)/mean(R2)<10^-3

where sigma(x) represents the RMSE of the metric x.

Quantitative result table on test stars

model	pixel star RMSE	sigma(e1) RMSE	sigma(e2) RMSE	sigma(R2)/ RMSE
parametric_cycle1	2.4318e-04	4.2143e-02	4.4921e-02	1.6890e-01
-	-	-	-	-
poly_cycle1	9.5582e-05	1.2334e-02	1.1746e-02	6.1628e-02
poly_cycle2	8.7014e-05	1.1821e-02	6.0005e-03	2.2907e-02
-	-	-	-	-
graph_poly_cycle1	8.5684e-05	1.3745e-02	6.7043e-03	3.3967e-02
graph_poly_cycle2	1.0122e-04	2.6365e-02	2.2866e-02	1.7144e-01

Example of a PSF estimation at x3 resolution and metric errors

[tobias-liaudat]

Solving this issue as it is old and now all the tests are done with under sampled and noisy stars.