AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

SuperDisco: Super-Class Discovery Improves Visual Recognition for the Long-Tail

Yingjun Du, Jiayi Shen, Xiantong Zhen, Cees G. M. Snoek

CVPR 2023 · Label-efficient and robust learning › Recognition and long-tail · approach: Decoupled and calibrated long-tail training · 18 citations (OpenAlex)

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add the discovered super-class graph and feature-refinement message passingfine-tuning baseline · CIFAR-100-LTaccuracy (%): 39.3 -> 45.2Improved
add prototype-graph meta-learning to SuperDisco (ablation)SuperDisco · CIFAR-100-LTaccuracy (%): 60.3 -> 68.5; 57.3 -> 63.1; 47.5 -> 58.3; 44.9 -> 53.8; 39.3 -> 47.5Improved
use a multi-level discovered super-class graph rather than one level (20) (ablation)Meta-SuperDisco · CIFAR-100-LT, imbalance ratio 10accuracy (%): 61.2 -> 69.1Improved
increase hierarchy from (4, 8, 16) to (4, 8, 16, 32) (ablation)Meta-SuperDisco · CIFAR-100-LT, imbalance ratio 10accuracy (%): 69.1 -> 68.5No clear effect
use five super-class levels (4, 8, 16, 32, 64) instead of four (4, 8, 16, 32) (ablation)Meta-SuperDisco · CIFAR-100-LT, imbalance ratio 10accuracy (%): 68.5 -> 66.9Hurt
use discovered super-classes rather than ground-truth oracle super-classes (ablation)Meta-SuperDisco · CIFAR-100-LT, imbalance ratio 10accuracy (%): 66.9 -> 69.1Improved
use a deeper, wider hierarchy rather than one level with 13 super-classes (ablation)Meta-SuperDisco · iNaturalist, Allaccuracy (%): 70.8 -> 72.3Improved
replace the super-class mechanism with a multi-layer perceptron (ablation)SuperDisco · ImageNet-LT, Allaccuracy (%): 57.1 -> 55.0Hurt
add the super-class mechanism to the baseline representation (ablation)ImageNet-LT · ImageNet-LT, Allaccuracy (%): 52.7 -> 55.0Improved
add Meta-SuperDisco to the long-tailed recognition systemlong-tailed classifiers · ImageNet-LT; Places-LT; iNaturalistAll accuracy (%): 56.2 -> 57.1; 39.2 -> 40.3; 72.8 -> 73.6Mixed
add Meta-SuperDisco to the long-tailed recognition systemlong-tailed classifiers · ImageNet-LT, Fewaccuracy (%): 41.5 -> 37.1Hurt
use Meta-SuperDisco for long-tailed recognitionlong-tailed classifiers · CIFAR-100-LTaccuracy (%): 68.8 -> 69.3; 57.6 -> 58.3; 52.0 -> 53.8Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy