Research map / Detection and segmentation / Weakly supervised detection and segmentation
Explanation-based Weakly-supervised Learning of Visual Relations with Graph Networks
Federico Baldassarre, Kevin Smith, Josephine Sullivan, Hossein Azizpour
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 |
|---|---|---|---|
| use human-object graph prior instead of uniform fully-connected graph (ablation) | explanation-based detector · HICO-DET Full (600) | mAP: 24.25 -> 28.04 | Improved |
| replace uniform prior with frequency-based relationship prior (ablation) | explanation-based detector · VRD predicate detection (GT objects) | recall@50: 27.3 -> 43.0 | Improved |
| replace uniform prior with frequency-based relationship prior (ablation) | explanation-based detector · VRD relationship detection (R-CNN objects) | recall@50: 5.3 -> 10.6 | Improved |
| replace uniform prior with frequency-based relationship prior (ablation) | explanation-based detector · VRD zero-shot predicate detection (GT objects) | recall@50: 13.7 -> 13.5 | No clear effect |
| give prior access to a few zero-shot triplets (ablation) | explanation-based detector · VRD zero-shot predicate detection (GT objects) | recall@50: 13.7 -> 20.5 | Improved |
| replace uniform prior with frequency prior (ablation) | explanation-based detector · UnRel relationship detection (R-CNN objects) | mAP: 14.9 -> 15.1 | No clear effect |
| use image-level predicate labels only versus fully supervised triplet labels | explanation-based detector · HICO-DET Full (600) | mAP: 19.40 -> 28.04 | Improved |
| weak predicate-only supervision versus weak triplet supervision (Peyre) | explanation-based detector · VRD predicate detection (GT objects) | recall@100: 46.8 -> 57.4 | Improved |
| weak predicate-only supervision versus weak triplet supervision (Peyre) | explanation-based detector · VRD relationship detection (R-CNN objects) | recall@50: 14.1 -> 10.6 | Hurt |
| change relevance computation (gradient x input, max(.,0), L1 or L2 norm) (ablation) | explanation step · validation split | relationship detection | No clear effect |
| use mean pooling instead of max pooling in the graph aggregation (ablation) | predicate classifier · VRD | recall@50 and recall@100 | Mixed |
| increase number N of explained predicates beyond 10-20 (ablation) | explanation step · VRD | recall@50 and recall@100 | No clear effect |
Closest papers by meaning
- Weakly-Supervised Learning of Visual Relations (ICCV 2017 · similarity 0.93)
- Localize, Assemble, and Predicate: Contextual Object Proposal Embedding for Visual Relation Detection (AAAI 2020 · similarity 0.92)
- PPR-FCN: Weakly Supervised Visual Relation Detection via Parallel Pairwise R-FCN (ICCV 2017 · similarity 0.91)
- Visual Relationship Detection with Language Priors (ECCV 2016 · similarity 0.91)
- One-Shot Learning for Long-Tail Visual Relation Detection (AAAI 2020 · similarity 0.91)
- MR-NET: Exploiting Mutual Relation for Visual Relationship Detection (AAAI 2019 · similarity 0.90)
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.
