Self-Paced Learning Strategy with Easy Sample Prior Based on Confidence for the Flying Bird Object Detection Model Training Journal: Arxiv

Zi-Wei Sun, Ze-Xi hua, Heng-Chao Li, Yan Li · 2024

Read Self-Paced Learning Strategy with Easy Sample Prior Based on Confidence for the Flying Bird Object Detection Model Training Journal: Arxiv by Zi-Wei Sun, Ze-Xi hua, Heng-Chao Li, Yan Li (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-06306.txt.

Description

Zi-Wei Sun and three co-authors propose a gentler way to train a bird-spotting model in this 2024 paper. Their earlier FBOD model detects flying birds in surveillance video, a one-class task where hard examples can drag training off course. The new strategy lets the model learn from easy examples first and only later face the hard ones.

The method, called SPL-ESP-BC, reworks self-paced learning in two ways. Its pace rule judges examples by confidence rather than by loss, which fits one-class detection better. A warm-up stage on easy samples also teaches the model to tell easy from hard before the full training begins. Trained this way, the bird detector gained about two points on a standard accuracy measure over plain training. It also beat other self-paced variants on overall detection quality.

Catalog record

CREATORZi-Wei Sun, Ze-Xi hua, Heng-Chao Li, Yan Li
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE18 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-06306

Browse the full library · About the Tower of Babel