AI Paper Atlas

Research map / Detection and segmentation / CNN features

Revisiting Multi-Task Learning with ROCK: a Deep Residual Auxiliary Block for Visual Detection

Taylor Mordan, Nicolas THOME, Gilles Henaff, Matthieu Cord

NeurIPS 2018 · Detection and segmentation › CNN features · approach: Local features and context in CNNs · 40 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 residual auxiliary block (ROCK) with auxiliary supervision to SSDSSD ResNet-50 · NYUv2 valmAP@0.5: 31.2 -> 37.6Improved
add auxiliary annotations via flat MTL heads (ablation)SSD ResNet-50 · NYUv2 valmAP@0.5: 31.2 -> 34.3Improved
add ROCK auxiliary block encoder/predictor without fusion over flat MTL (ablation)flat MTL SSD · NYUv2 valmAP@0.5: 34.3 -> 35.7Improved
add feature merging (decoder and fusion) (ablation)ROCK w/o fusion · NYUv2 valmAP@0.5: 35.7 -> 37.6Improved
add scene classification supervision (ablation)ROCK · NYUv2 valmAP@0.5: 34.0 -> 37.6Improved
add normal supervision (ablation)ROCK · NYUv2 valmAP@0.5: 33.2 -> 37.6Improved
add depth supervision (ablation)ROCK · NYUv2 valmAP@0.5: 35.1 -> 37.6Improved
enable auxiliary losses (vs weights set to 0) (ablation)ROCK · NYUv2 valmAP@0.5: 30.6 -> 37.6Improved
element-wise product instead of addition for fusion (ablation)ROCK depth-only · NYUv2 valmAP@0.5: 30.9 -> 32.3Improved
pre-train on synthetic MLT dataset then fine-tuneROCK (DNS) · NYUv2 testmAP@0.5: 39.8 -> 46.8Improved
apply ROCK with disparity auxiliary taskSSD ResNet-50 · CityScapes valmAP@0.5: 42.8 -> 44.0Improved
train on 70% of data with ROCK vs baseline on 100%ROCK · NYUv2 valmAP@0.5: 31.2 vs 32.8Mixed

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