Open Access

Table 4

Performance comparison in terms of average F1 score and the average of the accuracy of five folds of cross-validation.

Model Bands Type of data Accuracy F1 score Num params
ConvEntion (Ours) ugriz Images 79.83 70.62 1.253M
CNN+GRU (Gómez et al. 2020) ugriz Images 66.39 63.22 1.993M
ConvEntion (Ours) g Images 76.89 63.20 1.253M
CNN+GRU (Gómez et al. 2020) g Images 63.67 61.00 1.992M
CNN+LSTM (Carrasco-Davis et al. 2019) ugriz Images 64.08 60.65 2.190M
CNN+LSTM (Carrasco-Davis et al. 2019) g Images 63.00 60.00 2.189M

SuperNNova (Bayes) (Möller & de Boissière 2020) ugriz Light curves 65.54 55.40
SITS-BERT (Yuan & Lin 2021) ugriz Light curves 67.43 51.60 0.596M
SCONE (CNN) (Qu et al. 2021) ugriz Light curves 62.57 50.43 22.2K
SuperNNova (RNN) (Möller & de Boissière 2020) ugriz Light curves 56.30 42.60
LSTM ugriz Light curves 55.24 40.33 60K

Notes. This table includes only experiments on a dataset with four classes.

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.