</> Code with Papers by Syntology

Papers › From Generalized zero-shot learning to long-tail with class descriptors

From Generalized zero-shot learning to long-tail with class descriptors

5 Apr 2020arXiv:2004.02235archive 2025-07-28

Dvir Samuel, Yuval Atzmon, Gal Chechik

Real-world data is predominantly unbalanced and long-tailed, but deep models struggle to recognize rare classes in the presence of frequent classes. Often, classes can be accompanied by side information like textual descriptions, but it is not fully clear how to use them for learning with unbalanced long-tail data. Such descriptions have been mostly used in (Generalized) Zero-shot learning (ZSL), suggesting that ZSL with class descriptions may also be useful for long-tail distributions. We describe DRAGON, a late-fusion architecture for long-tail learning with class descriptors. It learns to (1) correct the bias towards head classes on a sample-by-sample basis; and (2) fuse information from class-descriptions to improve the tail-class accuracy. We also introduce new benchmarks CUB-LT, SUN-LT, AWA-LT for long-tail learning with class-descriptions, building on existing learning-with-attributes datasets and a version of Imagenet-LT with class descriptors. DRAGON outperforms state-of-the-art models on the new benchmark. It is also a new SoTA on existing benchmarks for GFSL with class descriptors (GFSL-d) and standard (vision-only) long-tailed learning ImageNet-LT, CIFAR-10, 100, and Places365.

PaperPDFCode

Code

dvirsamuel/DRAGON officialmentioned in papermentioned on GitHubtf report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Few-Shot LearningGeneralized Few-Shot LearningGeneralized Zero-Shot LearningLong-tail LearningLong-tail learning with class descriptorsZero-Shot Learning

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Generalized Few-Shot Learning AwA2 DRAGON Per-Class Accuracy (1-shot) 67.1 #4 of 6 Archive leaderboard report
Generalized Few-Shot Learning AwA2 DRAGON Per-Class Accuracy (10-shots) 81.9 #4 of 6 Archive leaderboard report
Generalized Few-Shot Learning AwA2 DRAGON Per-Class Accuracy (2-shots) 69.1 #4 of 6 Archive leaderboard report
Generalized Few-Shot Learning AwA2 DRAGON Per-Class Accuracy (20-shots) 83.3 #4 of 6 Archive leaderboard report
Generalized Few-Shot Learning AwA2 DRAGON Per-Class Accuracy (5-shots) 76.7 #4 of 6 Archive leaderboard report
Generalized Few-Shot Learning SUN DRAGON Per-Class Accuracy (1-shot) 41.0 #1 of 5 Archive leaderboard report
Generalized Few-Shot Learning SUN DRAGON Per-Class Accuracy (10-shots) 48.2 #1 of 5 Archive leaderboard report
Generalized Few-Shot Learning SUN DRAGON Per-Class Accuracy (2-shots) 43.8 #1 of 5 Archive leaderboard report
Generalized Few-Shot Learning SUN DRAGON Per-Class Accuracy (5-shots) 46.7 #1 of 5 Archive leaderboard report
Long-tail Learning CIFAR-10-LT (ρ=10) smDRAGON Error Rate 11.84 #38 of 50 Archive leaderboard report
Long-tail Learning CIFAR-10-LT (ρ=100) smDRAGON Error Rate 20.37 #22 of 28 Archive leaderboard report
Long-tail Learning CIFAR-100-LT (ρ=10) smDRAGON Error Rate 41.23 #28 of 31 Archive leaderboard report
Long-tail Learning CIFAR-100-LT (ρ=100) smDRAGON Error Rate 56.50 #56 of 66 Archive leaderboard report
Long-tail Learning ImageNet-LT smDRAGON Top-1 Accuracy 42.0 #60 of 69 Archive leaderboard report
Long-tail Learning Places-LT smDRAGON Top-1 Accuracy 38.1 #24 of 29 Archive leaderboard report
Long-tail learning with class descriptors AWA-LT DRAGON + Bal'Loss Long-Tailed Accuracy 92.2 #1 of 5 Archive leaderboard report
Long-tail learning with class descriptors AWA-LT DRAGON + Bal'Loss Per-Class Accuracy 76.2 #1 of 5 Archive leaderboard report
Long-tail learning with class descriptors AWA-LT DRAGON Long-Tailed Accuracy 94.1 #2 of 5 Archive leaderboard report
Long-tail learning with class descriptors AWA-LT DRAGON Per-Class Accuracy 74.1 #2 of 5 Archive leaderboard report
Long-tail learning with class descriptors CUB-LT DRAGON + Bal'Loss Long-Tailed Accuracy 66.5 #1 of 5 Archive leaderboard report
Long-tail learning with class descriptors CUB-LT DRAGON + Bal'Loss Per-Class Accuracy 60.1 #1 of 5 Archive leaderboard report
Long-tail learning with class descriptors CUB-LT DRAGON Long-Tailed Accuracy 67.7 #2 of 5 Archive leaderboard report
Long-tail learning with class descriptors CUB-LT DRAGON Per-Class Accuracy 57.8 #2 of 5 Archive leaderboard report
Long-tail learning with class descriptors ImageNet-LT-d DRAGON + Bal'Loss Per-Class Accuracy 53.5 #1 of 5 Archive leaderboard report
Long-tail learning with class descriptors ImageNet-LT-d DRAGON Per-Class Accuracy 51.2 #2 of 5 Archive leaderboard report
Long-tail learning with class descriptors SUN-LT DRAGON + Bal'Loss Long-Tailed Accuracy 38.5 #1 of 5 Archive leaderboard report
Long-tail learning with class descriptors SUN-LT DRAGON + Bal'Loss Per-Class Accuracy 36.1 #1 of 5 Archive leaderboard report
Long-tail learning with class descriptors SUN-LT DRAGON Long-Tailed Accuracy 40.4 #2 of 5 Archive leaderboard report
Long-tail learning with class descriptors SUN-LT DRAGON Per-Class Accuracy 34.8 #2 of 5 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections