Research map / Detection and segmentation / Weakly supervised detection and segmentation
TriLite: Efficient Weakly Supervised Object Localization with Universal Visual Features and Tri-Region Disentanglement
Arian Sabaghi, Jose Oramas
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 three-channel heatmap and adversarial loss together (vs binary, no adversarial loss) (ablation) | TriLite with TriHead (DINOv2 ViT-S/14) · CUB-200-2011 | GT-known localization accuracy: 97.6 -> 98.5 | Improved |
| add three-channel heatmap and adversarial loss together (vs binary, no adversarial loss) (ablation) | TriLite with TriHead (DINOv2 ViT-S/14) · ImageNet-1K | GT-known localization accuracy: 76.5 -> 77.9 | Improved |
| add adversarial background loss to binary heatmap (ablation) | TriLite with TriHead (DINOv2 ViT-S/14) · CUB-200-2011 | GT-known localization accuracy: 97.6 -> 97.5 | No clear effect |
| add adversarial background loss to binary heatmap (ablation) | TriLite with TriHead (DINOv2 ViT-S/14) · ImageNet-1K | GT-known localization accuracy: 76.5 -> 77.2 | Improved |
| use three-channel instead of binary heatmap, without adversarial loss (ablation) | TriLite with TriHead (DINOv2 ViT-S/14) · CUB-200-2011 | GT-known localization accuracy: 97.6 -> 97.0 | Hurt |
| use three-channel instead of binary heatmap, without adversarial loss (ablation) | TriLite with TriHead (DINOv2 ViT-S/14) · ImageNet-1K | GT-known localization accuracy: 76.5 -> 77.4 | Improved |
| add adversarial loss to three-channel heatmap (ablation) | TriLite with TriHead (DINOv2 ViT-S/14) · CUB-200-2011 | Top-1 localization accuracy: 85.0 -> 87.3 | Improved |
| add adversarial loss to three-channel heatmap (ablation) | TriLite with TriHead (DINOv2 ViT-S/14) · ImageNet-1K | Top-1 localization accuracy: 65.0 -> 65.5 | Improved |
| use TriHead instead of LOST localization on frozen DINO ViT-S/16 (ablation) | TriLite (DINO) · CUB-200-2011 | GT-known localization accuracy: 89.7 -> 95.9 | Improved |
| use TriHead instead of TokenCut on frozen DINO ViT-S/16 (ablation) | TriLite (DINO) · ImageNet-1K | GT-known localization accuracy: 66.8 -> 73.2 | Improved |
| swap supervised DeiT-S backbone for self-supervised DINO backbone | TriLite · CUB-200-2011 | Top-1 localization accuracy: 69.2 -> 76.3 | Improved |
| swap supervised DeiT-S backbone for self-supervised DINO backbone | TriLite · ImageNet-1K | GT-known localization accuracy: 71.7 -> 73.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- TokenCut — TokenCut: Segmenting Objects in Images and Videos With Self-Supervised Transformer and Normalized Cut (TPAMI 2023) · better on 2 of 2 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
- Weakly Supervised Object Localization via Transformer with Implicit Spatial Calibration (ECCV 2022 · similarity 0.91)
- Weakly Supervised Object Localization and Detection: A Survey (TPAMI 2021 · similarity 0.90)
- Evaluation for Weakly Supervised Object Localization: Protocol, Metrics, and Datasets (TPAMI 2022 · similarity 0.89)
- Rethinking the Route Towards Weakly Supervised Object Localization (CVPR 2020 · similarity 0.89)
- Shallow Feature Matters for Weakly Supervised Object Localization (CVPR 2021 · similarity 0.89)
- Recognize Any Regions (NeurIPS 2024 · similarity 0.87)
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.
