AI Paper Atlas

Research map / Detection and segmentation / CNN features

Semantic Fisher Scores for Task Transfer: Using Objects to Classify Scenes

Mandar Dixit, Yunsheng Li, Nuno M. Vasconcelos

TPAMI 2019 · Detection and segmentation › CNN features · approach: Local features and context in CNNs

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
embed semantics in natural parameter space (log) before GMM-FVSMN-FV · MIT Indoor and SUNaccuracy (%): around 9%Improved
use Gaussian scaling instead of Fisher scaling F^-1/2 (ablation)DMM-FV · MIT Indooraccuracy (%): 58.8 -> 67.1Improved
use Gaussian scaling instead of Fisher scaling F^-1/2 (ablation)GMM-FV · MIT Indooraccuracy (%): 57.7 -> 68.5Improved
change assignment function q to h (ablation)GMM-FV · MIT Indooraccuracy (%): 68.7 -> 68.5No clear effect
use nu(2) pre-softmax embedding instead of nu(1) (ablation)GMM-FV · MIT Indooraccuracy (%): 67.7 -> 68.5Improved
use nu(2) instead of nu(3) (ablation)GMM-FV · SUNaccuracy (%): 50.47 -> 51.17Improved
use natural-parameter nu(2) instead of square-root embeddingGMM-FV · MIT Indooraccuracy (%): 58.95 -> 68.5Improved
use natural-parameter nu(2) instead of von-Mises embeddingGMM-FV · SUNaccuracy (%): 46.1 -> 51.17Improved
add log to fc7 features (ablation)fc7-FV · MIT Indooraccuracy (%): 65.1 -> 58Hurt
add log to nu(2) features (ablation)nu(2)-FV · MIT Indooraccuracy (%): 68.5 -> 60.97Hurt
use full-covariance MFA-FS instead of variance GMM-FVFV pooling of CNN features · MIT Indooraccuracy (%): 60.7Improved
random instead of MFA-FS initialization (ablation)MFAFSNet · MIT Indooraccuracy (%): 2-3%Hurt

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