Table A.1.
Comparison of the performances of the detection methods.
| Method | Input | Output | Param. | Acc. | Pros | Cons |
|---|---|---|---|---|---|---|
| Convolutional ResNet | 6 filtres ugrizy 512×512 images | (x, y) pixel coordinates | 830,882 | 81% | – Good accuracy | – No flag for bad detections |
| – Straightforward | – Gets confused by the presence of multiple candidates | |||||
| – No redshift dependencies | – Training is computationally expensive | |||||
| – Difficulties to detect sources with low-SNR and deblending issues | ||||||
| Autoencoder ResNet | 6 filtres ugrizy 512×512 images | Probability map | 531,719 | 95% | – Very good accuracy and resolution | – Difficulties to detect sources with low-SNR and deblending issues |
| – Provides a confidence level for each detection | ||||||
| – Can detect multiple BCG candidates | ||||||
| – No redshift dependencies | ||||||
| Red sequence algorithm | Photometric catalogues in at least two filtres | BCG ID in the catalogue | 70% | – Physically motivated | – Lower accuracy | |
| – Requires pre-processing of photometric catalogues | ||||||
| – Difficulties to detect sources with low-SNR and deblending issues | ||||||
| – Photometry less reliable at high redshifts | ||||||
| – Need redder filtres to contain the 4000Å at higher redshifts | ||||||
Notes. Results from the three different detection methods used in this paper: the convolutional ResNet detailed in Sect. 4.1, the autoencoder ResNet described in Sect. 4.2, and the red sequence based algorithm used as benchmark and discussed further in Sect. 6.1. The columns are, from left to right: the method used, the input and the output of each method, the number of parameters in the NN, the accuracy defined as the number of good detections over the total number of clusters in the evaluation sample, and finally the pros and cons of each method.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.