Research map / Detection and segmentation / Open-set and few-shot detection
OLAF: A Plug-and-Play Framework for Enhanced Multi-object Multi-part Scene Parsing
Pranav Gupta, Rishubh Singh, Pradeep Shenoy, Ravi Kiran Sarvadevabhatla
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add OLAF to FLOAT | FLOAT · Pascal-Part-201 | mIoU: 46.9 -> 50.7 (ViT-H) | Improved |
| add OLAF to Segformer MiT-B2 | Segformer · PartImageNet val | mean accuracy: 71.62 -> 79.10 | Improved |
| add OLAF to Deeplabv3, BSANet and GMNet | CNN part-segmentation models · Pascal-Part-58 | mIoU: 54.3 -> 59.0 (Deeplabv3) | Improved |
| add LDF module alone (ablation) | FLOAT ViT-H · Pascal-Part-201 | mIoU: 37.7 -> 38.8 | Improved |
| add LDF on top of edge and fg/bg channels (ablation) | FLOAT ViT-H · Pascal-Part-201 | mIoU: 39.2 -> 40.9 | Improved |
| add LDF improves small parts more than input channels (ablation) | FLOAT ViT-H · Pascal-Part-201 | mIoU small: 24.0 -> 25.7 | Improved |
| use SAM foreground map instead of object segmentation network (ablation) | OLAF · Pascal-Part-201 | mIoU: 40.9 -> 40.5 | Hurt |
| use Canny or EDTER edges instead of HED (ablation) | OLAF · Pascal-Part-201 | mIoU: 40.9 -> 39.0 (Canny), 39.5 (EDTER) | Hurt |
| add monocular depth map as extra input channel (ablation) | OLAF · Pascal-Part-201 | mIoU: 40.9 -> 40.7 (Marigold), 40.8 (Depth Anything) | No clear effect |
| use average-of-RGB weight init plus warm-up instead of random or other adaptation schemes (ablation) | OLAF 5-channel input layer · Pascal-Part-201 | mIoU: 35.2 -> 40.9 (vs Random-5) | Improved |
| increase inference resolution from 770 to 1024 (ablation) | FLOAT · Pascal-Part-58 | mIoU: 61.0 -> 57.4 | Hurt |
| train FLOAT at higher resolution 770 instead of 513 (ablation) | FLOAT · Pascal-Part-58 | mIoU: 60.8 -> 60.5 | No clear effect |
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.
- FLOAT — FLOAT: Factorized Learning of Object Attributes for Improved Multi-Object Multi-Part Scene Parsing (CVPR 2022) · better on 3 of 3 shared results
- Segformer — SegFormer: A Topic Segmentation Model with Controllable Range of Attention (AAAI 2023) · better on 1 of 1 shared result
- GMNet — GMNet: Graph Matching Network for Large Scale Part Semantic Segmentation in the Wild (ECCV 2020) · better on 1 of 1 shared result
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
- FLOAT: Factorized Learning of Object Attributes for Improved Multi-Object Multi-Part Scene Parsing (CVPR 2022 · similarity 0.92)
- From Pixels to Objects: A Hierarchical Approach for Part and Object Segmentation Using Local and Global Aggregation (ECCV 2024 · similarity 0.86)
- Going Denser with Open-Vocabulary Part Segmentation (ICCV 2023 · similarity 0.86)
- Scene Parsing Through ADE20K Dataset (CVPR 2017 · similarity 0.86)
- Fine-Grained Image-Text Correspondence with Cost Aggregation for Open-Vocabulary Part Segmentation (CVPR 2025 · similarity 0.85)
- LangHOPS: Language Grounded Hierarchical Open-Vocabulary Part Segmentation (NeurIPS 2025 · similarity 0.85)
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.
