Table B.1
Summary of the ML and DL methods: Name or architecture, number of trainable parameters, whether one network is trained in the whole dataset or four networks are trained (one for each redshift bin), the size of the image in pixels, the size of the image in kpc, the scaling used in the images, the threshold used for the binary classification task, the kind of data augmentation used during training, any extra data that was used, whether the method performed the multi-class classification task, and a paper reference for the method.
| Method-1 (RF) | Method-2 (Swin) | Method-3 (Zoobot) | Method-4 (CNN1) | Method-5 (CNN2) | Method-6 (CNN3) | |
|---|---|---|---|---|---|---|
| Architecture | RF | S winTrans former | Pretrained-CNN | CNN | CNN | CNN |
| (EfficientnetB0) | (4conv+2dense) | (4conv+2dense) | (3conv+1dense) | |||
| Trainable | 2 × 108 (binary task) | 2× 105 | 5.3 × 106 | 0.1 <z< 0.31: 4.6 × 106 | 8.1 × 106 | 6.1 × 10s |
| parameters | 8 × 107 (multiclass task) | 0.31 < z < 0.52: 3.7 × 106 0.52 <z< 0.76: 3.0× 106 0.76 <z< 1.0: 2.5 × 106 | ||||
| 4 bins together? | Yes | Yes | Yes | No | Yes | Yes |
| Size (pixels) | 310/192/160/128 | 224 | 200 | 310/192/160/128 | 128 | 100 |
| Size (kpc) | 160 | 56/93/112/140 | 100 | 80 | 160 | 160 |
| Scaling | None | linear | arcsinh | arcsinh+clip | linear | AsinStrech |
| [0,1] | [0,1] | [0,1] | +Percentile(97) | |||
| Threshold | 0.5 | – | 0.56a | 0.5 | 0.51b | 0.51a |
| Data augm. | – | Rotation (0°, 90°, 180°, or 270°), horizontal flip, vertical flip | Rotation, horizontal flip, vertical flip | Rotation (0°– 90°), horizontal flip, vertical flip zoom [0.7,1.1] | Rotation (–45°– 45°) horizontal flip, vertical flip, zoom [0.75,1.3], xy translation [–0.05, 0.05) | |
| Extra data | No | Yes (Pretrained ImageNet-lk) | Yes (Galaxy Zoo) | No | No | No |
| multi-class | Yes | Yes | Yes | Yes | No | No |
| reference | Guzmán-Ortega et al. (2023) | Minghao et al. (2021) | Walmsley et al. (2023) | Bickley et al. (2021) | Chudy et al. in prep. | Walmsley et al. (2019) |
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.