AI Paper Atlas

Research map / Video / Action recognition

Compound Memory Networks for Few-shot Video Classification

Linchao Zhu, Yi Yang

ECCV 2018 · Video › Action recognition · approach: Contrastive video representation learning · 254 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 constituent key memory (matrix keys) instead of single-vector key (ablation)Plain CMN · Kinetics 5-way 1-shotaccuracy: 57.3 -> 60.5Improved
add constituent key memory (matrix keys) instead of single-vector key (ablation)Plain CMN · Kinetics 5-way 5-shotaccuracy: 76.0 -> 78.9Improved
use multi-saliency embedding instead of LSTM embedding (ablation)LSTM-emb · Kinetics 5-way 1-shotaccuracy: 57.6 -> 60.5Improved
use CMN instead of Matching NetworksMatchingNet · Kinetics 5-way 5-shotaccuracy: 74.6 -> 78.9Improved
fine-tune ResNet-50 on meta-training classes before nearest neighbour (ablation)Nearest-pretrain · Kinetics 5-way 1-shotaccuracy: 51.1 -> 48.2Hurt
increase memory size from 64 to 512 slots (ablation)CMN · Kinetics 5-way 1-shot (meta-val)accuracy: 52.0 -> 55.1Improved
increase memory size from 512 to 2048 slots (ablation)CMN · Kinetics 5-way 1-shot (meta-val)accuracy: 55.1 -> 55.0No clear effect
use 5 multi-saliency descriptors instead of 1 (ablation)CMN · Kinetics 5-way 1-shot (meta-val)accuracy: 53.7 -> 55.1Improved
increase descriptors from 5 to 10 (ablation)CMN · Kinetics 5-way 1-shot (meta-val)accuracy: 55.1 -> 53.2Hurt
increase number of classes per episode from 5 to 8 way (ablation)CMN · Kinetics 1-shot (meta-val)accuracy: 55.0 -> 46.0Hurt
add one more support example per class (1-shot to 2-shot) (ablation)CMN · Kinetics 5-wayaccuracy: 60.5 -> 70.0Improved

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