Issues

• sequential_train_val: sequence id present in both wildfire/ and fp/ pools (train split, v4.1.0)

  pyronear/pyro-dataset#39 · Filter by author Chouffe opened on Aug 14, 2026

• Test sets are not reproducible across rebuilds: 51% of sequential_test negatives churned

  pyronear/pyro-dataset#33 · Filter by author Chouffe opened on Aug 14, 2026

• FP selection ranks annotator sequences on a confidence column their labels do not have

  pyronear/pyro-dataset#29 · Filter by author Chouffe opened on Aug 14, 2026
  1

• Two fp sequences share 29 byte-identical frames across val and test (latent leakage)

  pyronear/pyro-dataset#25 · Filter by author Chouffe opened on Aug 13, 2026

• One sequence is registered in both the wildfire and fp pools, with colliding image filenames

  pyronear/pyro-dataset#24 · Filter by author Chouffe opened on Aug 13, 2026

• Raw label files carry stray class ids; 140 lines are silently dropped downstream

  pyronear/pyro-dataset#22 · Filter by author Chouffe opened on Aug 13, 2026

• Publish pyro-SDIS 2025 on Hugging Face

  pyronear/pyro-dataset#14 · Filter by author fe51 opened on Oct 3, 2025