AI Paper Atlas

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

Product Quantization Network for Fast Image Retrieval

Tan Yu, Junsong Yuan, Chen Fang, Hailin Jin

ECCV 2018 · Label-efficient and robust learning › Recognition and long-tail · approach: Correspondence and causal representation learning · 71 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
use asymmetric triplet loss instead of triplet loss (ablation)PQN · CIFAR-10 (3CNet, 36 bits)mAP: 0.778 -> 0.787Improved
use asymmetric triplet loss instead of cross-entropy loss (ablation)PQN · CIFAR-10 (3CNet, 36 bits)mAP: 0.768 -> 0.787Improved
use asymmetric triplet loss instead of cross-entropy loss (ablation)PQN · CIFAR-10 (3CNet, 24 bits)mAP: 0.771 -> 0.782Improved
use learned soft product quantization instead of unsupervised PQ on triplet featuresPQN · CIFAR-10 (3CNet, 4 bits)mAP: 0.503 -> 0.574Improved
use PQN instead of full-precision triplet featuresPQN · CIFAR-10 (3CNet, 32 bits)mAP: 0.779 -> 0.786Improved
use PQN instead of DPQ (no extra fully-connected layers)PQN · CIFAR-10 (3CNet, 24 bits)mAP: 0.692 -> 0.782Improved
use PQN instead of SUBICPQN · CIFAR-10 (3CNet, 24 bits)mAP: 0.672 -> 0.782Improved
increase M from 1/2 to 4 subvectors (ablation)PQN · CIFAR-10 (3CNet)mAPMixed
increase B from 9 to 12 bits with M = 4 (ablation)PQN · CIFAR-10 (3CNet)mAPHurt
increase alpha from 1 to 5 (softness of quantization) (ablation)PQN · CIFAR-10 (3CNet)mAPImproved
increase alpha from 20 to 40 (ablation)PQN · CIFAR-10 (3CNet)mAPHurt
reduce code length to 4 bits with M = 1 (ablation)PQN · CIFAR-10 (AlexNet)mAP: 0.945 at 4 bits vs DSDH 0.935 at 16 bitsNo clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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