AI Paper Atlas

Research map / Detection and segmentation / CNN features

SALISA: Saliency-Based Input Sampling for Efficient Video Object Detection

Babak Ehteshami Bejnordi, Amirhossein Habibian, Fatih Porikli, Amir Ghodrati

ECCV 2022 · Detection and segmentation › CNN features · approach: RGB-D salient object detection · 20 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 saliency-based input resamplingEfficientDet-D0 · UA-DETRACmAP: 51.3 -> 54.2Improved
add saliency-based input resamplingEfficientDet-D2 · UA-DETRACmAP: 59.4 -> 61.2Improved
add saliency-based input resamplingEfficientDet-D2 · ImageNet-VIDmAP: 72.5 -> 74.5Improved
add saliency-based input resamplingEfficientDet-D1 · UA-DETRAC small objectssmall-object mAP: 8.4 -> 14.9Improved
add saliency-based input resamplingEfficientDet-D1 · UA-DETRAC medium objectsmAP: 54.8 -> 58.2Improved
add saliency-based input resamplingEfficientDet-D3 · UA-DETRAC large objectsmAP: 78.5 -> 78.0No clear effect
add the regularization loss to the TPS-STN resampler (ablation)SALISA with EfficientDet-D1 · UA-DETRACmAP: 57.6 -> 61.4Improved
use SALISA resampler instead of learning to zoom (ablation)SALISA with EfficientDet-D1 · UA-DETRACmAP: 47.7 -> 61.4Improved
increase TPS control points from 256 to 1024 (ablation)SALISA with EfficientDet-D0 · UA-DETRACmAP: 54.2 -> 45.7Hurt
lower area threshold alpha from 2.0 to 0.5 (ablation)SALISA with EfficientDet-D1 · UA-DETRAC small objectssmall-object mAP: 13.1 -> 14.9Improved
lower alpha from 0.5 to 0.1 (ablation)SALISA with EfficientDet-D1 · UA-DETRAC small objectssmall-object mAP: 14.9 -> 13.4Hurt
lower saliency threshold tau from 0.5 to 0.3 (ablation)SALISA with EfficientDet-D0 · ImageNet-VIDmAP: 70.2 -> 70.4No clear effect

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