EvalNorm: Estimating Batch Normalization Statistics for Evaluation
This addresses a practical problem for deep learning practitioners using BN with small batch sizes, offering an incremental improvement to existing methods.
The paper tackled the performance degradation of batch normalization (BN) in deep learning models when trained with small minibatches, proposing EvalNorm to estimate corrected normalization statistics for evaluation, which resulted in absolute gains of up to 6.18% on ImageNet and 1.5 to 7.0 points on COCO.
Batch normalization (BN) has been very effective for deep learning and is widely used. However, when training with small minibatches, models using BN exhibit a significant degradation in performance. In this paper we study this peculiar behavior of BN to gain a better understanding of the problem, and identify a cause. We propose 'EvalNorm' to address the issue by estimating corrected normalization statistics to use for BN during evaluation. EvalNorm supports online estimation of the corrected statistics while the model is being trained, and does not affect the training scheme of the model. As a result, EvalNorm can also be used with existing pre-trained models allowing them to benefit from our method. EvalNorm yields large gains for models trained with smaller batches. Our experiments show that EvalNorm performs 6.18% (absolute) better than vanilla BN for a batchsize of 2 on ImageNet validation set and from 1.5 to 7.0 points (absolute) gain on the COCO object detection benchmark across a variety of setups.