A black-box active learning method selects 30% of camera-trap data and outperforms full-data training on the SAWIT benchmark in a single-run evaluation.
This is enabled by incorporating both un- certainty and diversity quantities in the sampling process
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
A model-agnostic active learning approach for animal detection from camera traps
A black-box active learning method selects 30% of camera-trap data and outperforms full-data training on the SAWIT benchmark in a single-run evaluation.