Research map / Detection and segmentation / Weakly supervised detection and segmentation
Adaptive Reconstruction Network for Weakly Supervised Referring Expression Grounding
Xuejing Liu, Liang Li, Shuhui Wang, Zheng-Jun Zha, Dechao Meng, Qingming Huang
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 adaptive reconstruction loss to language reconstruction (ablation) | ARN · RefCLEF test | accuracy IoU>0.5 (%): 21.86 -> 25.35 | Improved |
| add attribute classification loss (ablation) | ARN (L_lan+L_adp) · RefCLEF test | accuracy IoU>0.5 (%): 25.35 -> 26.19 | Improved |
| add attribute classification loss, lan weight 1 (ablation) | ARN · RefCLEF val | accuracy IoU>0.5 (%): 22.34 -> 25.35 | Improved |
| raise language reconstruction weight gamma from 10 to 30 (ablation) | ARN · RefCLEF val | accuracy IoU>0.5 (%): 24.34 -> 26.19 | Improved |
| raise language reconstruction weight gamma from 30 to 40 (ablation) | ARN · RefCLEF val | accuracy IoU>0.5 (%): 26.19 -> 25.53 | Hurt |
| raise adaptive visual reconstruction weight alpha from 0.001 to 0.01 (ablation) | ARN · RefCLEF val | accuracy IoU>0.5 (%): 24.14 -> 21.83 | Hurt |
| raise gamma from 5 to 15 (ablation) | ARN · RefCOCO testA | accuracy IoU>0.5 (%): 36.01 -> 27.13 | Hurt |
| replace adaptive loss with language reconstruction loss (ablation) | ARN (L_adp+L_att vs L_lan+L_att) · RefCOCO testB | accuracy IoU>0.5 (%): 29.09 -> 36.47 | Improved |
| replace language loss with adaptive loss (ablation) | ARN (L_lan+L_att vs L_adp+L_att) · RefCOCO testA | accuracy IoU>0.5 (%): 35.27 -> 36.43 | Mixed |
| use detected Faster R-CNN proposals instead of ground-truth boxes (ablation) | ARN · RefCOCO val | accuracy IoU>0.5 (%): 34.26 -> 32.17 | Hurt |
| full ARN versus previous best KAC Net | ARN · RefCLEF test | accuracy IoU>0.5 (%): 15.83 -> 26.19 | Improved |
| language reconstruction only versus previous best | ARN (L_lan) · RefCLEF test | accuracy IoU>0.5 (%): 15.83 -> 21.86 | Improved |
Closest papers by meaning
- Entity-enhanced Adaptive Reconstruction Network for Weakly Supervised Referring Expression Grounding (TPAMI 2022 · similarity 0.96)
- Discriminative Triad Matching and Reconstruction for Weakly Referring Expression Grounding (TPAMI 2021 · similarity 0.93)
- Knowledge Aided Consistency for Weakly Supervised Phrase Grounding (CVPR 2018 · similarity 0.91)
- Referencing Where to Focus: Improving Visual Grounding with Referential Query (NeurIPS 2024 · similarity 0.89)
- Hierarchical Alignment-enhanced Adaptive Grounding Network for Generalized Referring Expression Comprehension (AAAI 2025 · similarity 0.88)
- Boosting Weakly Supervised Referring Image Segmentation via Progressive Comprehension (NeurIPS 2024 · similarity 0.88)
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.
