AI Paper Atlas

Research map / Detection and segmentation / CNN features

SOLO: Segmenting Objects by Locations

Xinlong Wang, Tao Kong, Chunhua Shen, Yuning Jiang, Lei Li

ECCV 2020 · Detection and segmentation › CNN features · approach: Boundary-aware instance segmentation · 932 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 FPN multi-level prediction (five pyramids) over best single-scale grid (ablation)SOLO · COCO val2017mask AP: 29.0 -> 35.8Improved
increase grid number from 12 to 24 (single-scale) (ablation)SOLO · COCO val2017mask AP: 27.2 -> 29.0Improved
increase grid number from 24 to 36 (single-scale) (ablation)SOLO · COCO val2017mask AP: 29.0 -> 28.6Hurt
add one CoordConv layer (ablation)SOLO · COCO val2017mask AP: 32.2 -> 35.8Improved
add second and third CoordConv layers (ablation)SOLO · COCO val2017mask AP: 35.8 -> 35.7 / 35.8No clear effect
use Focal Loss instead of BCE for mask loss (ablation)SOLO · COCO val2017mask AP: 30.0 -> 31.6Improved
use Dice Loss instead of Focal Loss for mask loss (ablation)SOLO · COCO val2017mask AP: 31.6 -> 35.8Improved
use Dice loss instead of BCE for mask loss (ablation)Mask R-CNN · COCO val2017mask AP: 36.2 -> 35.3Hurt
increase mask head depth from 4 to 7 (ablation)SOLO · COCO val2017mask AP: 34.6 -> 35.8Improved
increase mask head depth from 7 to 8 (ablation)SOLO · COCO val2017mask AP: 35.8 -> 35.8No clear effect
change category-branch alignment (interpolation, adaptive-pool, region-grid-interpolation) (ablation)SOLO · COCO val2017mask AP: +/- 0.1 APNo clear effect
decouple the S^2 mask channels into X and Y tensors (ablation)SOLO · COCO test-dev / val2017mask AP: 0.6 AP gainsImproved

Later papers that compared with this method

7 later accepted papers report SOLO as a baseline in their results tables; 5 of them report a better number on most of the results they share with it.

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 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