OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Class Balanced Loss

Class Balanced Loss is a Python and TensorFlow implementation of effective-number-based class-balanced loss for training classifiers on long-tailed datasets. The repository provides CIFAR training and evaluation scripts, along with a TPU workflow for ImageNet and iNaturalist experiments.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Class Balanced Loss — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use it when investigating class imbalance in image-classification training and you can work with the repository's listed Python and TensorFlow versions or adapt its code.

Who it is for

  • Researchers comparing loss functions for long-tailed classification.
  • Machine-learning practitioners adapting class-balanced weighting in TensorFlow image-classification training.

Documented uses

  • Train and evaluate on original or long-tailed CIFAR using the provided scripts.
  • Compare standard training and effective-number-based class balancing on long-tailed CIFAR.
  • Prepare and run the documented TPU training workflows for ImageNet and iNaturalist datasets.
  • Inspect class distributions and effective sample numbers using the included notebook.
Before you choose

Requirements & limits

The public GitHub repository is MIT-licensed. The README lists Python 3.6 and TensorFlow 1.14 dependencies.

  • The documented CIFAR workflows use shell scripts and .tfrecords data.
  • The larger-scale TPU workflow requires dataset preparation, TFRecord conversion, Google Cloud Storage, and updates to image normalization and input size.
  • The repository lists TensorFlow 1.14, so users should account for that dependency when planning an environment.
  • The repository includes links to pretrained ResNet-50 models and TensorBoard logs for ImageNet 2012 and iNaturalist 2018.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Run the long-tailed CIFAR class-balanced training script

Use the repository's provided script for training and evaluation on long-tailed CIFAR with the proposed class-balanced loss.

Before you start

  • Python 3.6 and TensorFlow 1.14, as listed by the repository.
  • Long-tailed CIFAR data in the repository's expected .tfrecords format; the README links a dataset prepared for the paper.

Steps

  1. Obtain the repository and the linked long-tailed CIFAR .tfrecords data.

  2. Run ./cifar_im_trainval_cb.sh, setting BETA to a non-zero value as specified in the README.

  3. To visualize results, run tensorboard --logdir=./results --port=6006.

Expected outcome

The provided training and evaluation script uses class-balanced loss on long-tailed CIFAR; TensorBoard can visualize results in `./results`.

Guide limitations

  • The README does not provide a complete environment setup procedure or specify hardware requirements.
  • The documented TensorBoard command covers visualization, not a guarantee of training results.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
17 October 2026
Original website title
richardaecn/class-balanced-loss · Official homepage title · retained in its source language
How we review sourcesSuggest a correction