Research map / Detection and segmentation / Weakly supervised detection and segmentation
HASSOD: Hierarchical Adaptive Self-Supervised Object Detection
Shengcao Cao, Dhiraj Joshi, Liangyan Gui, Yu-Xiong Wang
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 HASSOD instead of CutLER | class-agnostic object detection · Objects365 val | Box AR: 35.8 -> 39.0 | Improved |
| Use HASSOD instead of CutLER | class-agnostic object detection and instance segmentation · LVIS v1.0 val | Mask AR: 20.2 -> 22.5 | Improved |
| Use HASSOD instead of CutLER | class-agnostic object detection and instance segmentation · SA-1B subset | Mask AR: 17.0 -> 26.0 | Improved |
| Add hierarchical level prediction (ablation) | HASSOD detector on MS-COCO+LVIS val2017 · MS-COCO+LVIS val2017 | Mask AR: 20.2 -> 20.6 | Improved |
| Add Mean Teacher self-training (ablation) | HASSOD detector with level prediction · MS-COCO+LVIS val2017 | Mask AR: 20.6 -> 22.1 | Improved |
| Adaptively increase teacher-target weight and decrease initial-label weight (ablation) | HASSOD with Mean Teacher · MS-COCO+LVIS val2017 | Mask AR: 22.1 -> 22.4 | Improved |
| Add post-processing to θmerge=0.1 clustering (ablation) | hierarchical adaptive clustering pseudo-labels · MS-COCO+LVIS val2017 pseudo-labels | Mask AR: 4.4 -> 4.1 | Hurt |
| Add post-processing to θmerge=0.2 clustering (ablation) | hierarchical adaptive clustering pseudo-labels · MS-COCO+LVIS val2017 pseudo-labels | Mask AR: 5.5 -> 5.3 | Hurt |
| Add post-processing to θmerge=0.4 clustering (ablation) | hierarchical adaptive clustering pseudo-labels · MS-COCO+LVIS val2017 pseudo-labels | Mask AR: 7.9 -> 7.8 | No clear effect |
| Ensemble pseudo-label masks from thresholds 0.1, 0.2, and 0.4 (ablation) | hierarchical adaptive clustering · MS-COCO+LVIS val2017 pseudo-labels | Mask AR: 7.8 -> 8.9 | Improved |
| Replace ViT-B/8 with ViT-B/16 for pseudo-label clustering at θmerge=0.2 (ablation) | hierarchical adaptive clustering · MS-COCO+LVIS val2017 pseudo-labels | Mask AR: 5.3 -> 3.8 | Hurt |
Later papers that compared with this method
1 later accepted paper reports HASSOD as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- MOD-UV: Learning Mobile Object Detectors from Unlabeled Videos (ECCV 2024 · better on 5 of 6 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
- UWSOD: Toward Fully-Supervised-Level Capacity Weakly Supervised Object Detection (NeurIPS 2020 · similarity 0.89)
- ALWOD: Active Learning for Weakly-Supervised Object Detection (ICCV 2023 · similarity 0.88)
- A Survey of Self-Supervised and Few-Shot Object Detection (TPAMI 2022 · similarity 0.88)
- Adaptive Hierarchical Representation Learning for Long-Tailed Object Detection (CVPR 2022 · similarity 0.88)
- Label, Verify, Correct: A Simple Few Shot Object Detection Method (CVPR 2022 · similarity 0.88)
- SimLTD: Simple Supervised and Semi-Supervised Long-Tailed Object Detection (CVPR 2025 · 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.
