Boosting High Resolution Image Classification with Scaling-up Transformers

TL;DR

This paper introduces a comprehensive pipeline for high-resolution image classification that leverages scaling-up transformers, transfer learning, data augmentation, and ensemble techniques to improve accuracy and robustness, achieving second place in a competitive challenge.

Contribution

It presents a holistic approach combining multiple strategies like domain analysis, transfer learning, and test-time augmentation specifically for high-resolution image classification with transformers.

Findings

Achieved second place in ICCV/CVPPA2023 challenge.

Demonstrated effectiveness of transfer learning and data augmentation.

Improved robustness through test-time augmentation and model ensembling.

Abstract

We present a holistic approach for high resolution image classification that won second place in the ICCV/CVPPA2023 Deep Nutrient Deficiency Challenge. The approach consists of a full pipeline of: 1) data distribution analysis to check potential domain shift, 2) backbone selection for a strong baseline model that scales up for high resolution input, 3) transfer learning that utilizes published pretrained models and continuous fine-tuning on small sub-datasets, 4) data augmentation for the diversity of training data and to prevent overfitting, 5) test-time augmentation to improve the prediction's robustness, and 6) "data soups" that conducts cross-fold model prediction average for smoothened final test results.