On Calibration of Scene-Text Recognition Models

Abstract

In this work, we study the problem of word-level confidence calibration for scene-text recognition (STR). Although the topic of confidence calibration has been an active research area for the last several decades, the case of structured and sequence prediction calibration has been scarcely explored. We analyze several recent STR methods and show that they are consistently overconfident. We then focus on the calibration of STR models on the word rather than the character level. In particular, we demonstrate that for attention based decoders, calibration of individual character predictions increases word-level calibration error compared to an uncalibrated model. In addition, we apply existing calibration methodologies as well as new sequence-based extensions to numerous STR models, demonstrating reduced calibration error by up to a factor of nearly 7. Finally, we show consistently improved accuracy results by applying our proposed sequence calibration method as a preprocessing step to beam-search.

Training Uncertainty-Aware Classifiers with Conformalized Deep Learning

“…Many methods have been developed to mitigate overconfidence in ML [20][21][22][23][24][25][26][27][28][29][30][31][32][33][34], for example by allowing an agnostic output, by suitable post-processing [7,[35][36][37][38][39][40], or through early stopping [41]. Additional relevant research includes that of [42][43][44][45][46][46][47][48][49][50][51][52], and these will provide us with informative benchmarks.…”

Context-Aware Selective Label Smoothing for Calibrating Sequence Recognition Model

“…Some authors considered the sequential characteristics and attempted to incorporate them into the sequential framework [19]. The varying sequential length is utilized as the setup basis of temperature values based on the temperature scaling technique [10,22,36]. Attention scores calculated during the sequence modeling are also utilized as the weights of temperature values [20].…”