# EvalNorm: Estimating Batch Normalization Statistics for Evaluation

## TL;DR

EvalNorm improves batch normalization during evaluation by estimating corrected statistics, significantly enhancing performance of models trained with small minibatches across tasks like image classification and object detection.

## Contribution

The paper introduces EvalNorm, a method for estimating better normalization statistics during evaluation, compatible with existing models and supporting online estimation.

## Key findings

- EvalNorm improves accuracy by 6.18% on ImageNet with batch size 2.
- EvalNorm yields 1.5 to 7.0 points gains on COCO detection.
- Supports online estimation without affecting training scheme.

## Abstract

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.