Table 3 Key hyperparameter settings used in the experiments.

From: A spatiotemporal transformer with cross-frame encoding and trajectory-aware decoding for multi-target fish tracking

Hyperparameter

Value

Batch size

16

Initial learning rate

1e-4

Optimizer

AdamW

Weight decay

0.05

LR scheduler

Cosine Annealing

Epochs

200

Input resolution

1280 \(\times\) 720

Flip probability

0.5

Multi-scale range

[0.8, 1.2]

Dataset Split

[7:1:2]