AI Paper Atlas

Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning

On the Importance of Distractors for Few-Shot Classification

Rajshekhar Das, Yu-Xiong Wang, José M. F. Moura

ICCV 2021 · Label-efficient and robust learning › Zero-/few-shot and meta-learning · approach: Embedding adaptation for few-shot classification · 32 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

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 contrastive finetuning (ConFT) on task samples and distractors to CE-trained priorCE-trained ResNet10 · Cars 5-shotaccuracy: 51.41 -> 61.53Improved
add ConFTCE-trained ResNet10 · CUB 5-shotaccuracy: 62.80 -> 70.53Improved
add multi-task cross-entropy loss to ConFTConFT · CUB 5-shotaccuracy: 70.53 -> 74.45Improved
add multi-task cross-entropy loss to ConFTConFT · Cars 1-shotaccuracy: 39.11 -> 37.36Hurt
add distractor negatives to task negatives (ablation)ConFT · Cars 5-shotaccuracy: 59.00 -> 60.58Improved
add task negatives alongside distractor negatives (ablation)ConFT · Cars 5-shotaccuracy: 57.42 -> 60.58Improved
add relative weighting of negatives (ablation)ConFT · Cars 5-shotaccuracy: 60.58 -> 61.53Improved
use stochastic pair construction rather than all task positives (ablation)ConFT · Cars 5-shotaccuracy: 61.01 -> 61.53Improved
enlarge distractor domain from 512 to 38400 images (ablation)ConFT · Cars 5-shotaccuracy: 61.16 -> 61.53No clear effect
use asymmetric rather than standard pair construction (ablation)ConFT · Cars 1-shotaccuracy: 37.09 -> 39.11Improved
add augmentation to task-specific samples (ablation)ConFT · Cars 5-shotaccuracy: 58.64 -> 61.53Improved
add ConFT finetuning to SimCLR priorSimCLR Conv4 · miniImageNet 5-shot unsupervisedaccuracy: 56.57 -> 60.02Improved

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