Research map / Label-efficient and robust learning / Recognition and long-tail
HARD-Net: Hardness-AwaRe Discrimination Network for 3D Early Activity Prediction
Tianjiao Li, Jun Liu, Wei Zhang, Lingyu Duan
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 HARD-Net adversarial hard-pair learning (ablation) | GCN backbone [34] · NTU RGB+D (cross-subject) | AUC over observation ratios: 66.91 -> 70.56 | Improved |
| add HARD-Net adversarial hard-pair learning (ablation) | GCN backbone [34] · NTU RGB+D (cross-view) | AUC over observation ratios: 75.50 -> 78.84 | Improved |
| add HARD-Net adversarial hard-pair learning (ablation) | GCN backbone [34] · FPHA | AUC over observation ratios: 72.17 -> 78.56 | Improved |
| add HARD-Net adversarial hard-pair learning (ablation) | GCN backbone [34] · FPHA | accuracy at 20% observation: 62.26 -> 71.83 | Improved |
| add HARD-Net adversarial hard-pair learning (ablation) | GCN backbone [34] · NTU RGB+D (cross-subject) | accuracy at 20% observation: 37.82 -> 42.39 | Improved |
| add HARD-Net adversarial hard-pair learning (ablation) | CNN backbone [13] · NTU RGB+D (cross-subject) | accuracy at 20% observation: 34.01 -> 35.86 | Improved |
| add HARD-Net adversarial hard-pair learning (ablation) | CNN backbone [13] · NTU RGB+D (cross-subject) | accuracy at 100% observation: 82.24 -> 82.98 | Improved |
| add ambiguous loss to generator (ablation) | HARD-Net · NTU RGB+D (cross-subject) | AUC over observation ratios: 67.9 -> 70.6 | Improved |
| add reconstruction (mean absolute error) loss to generator (ablation) | HARD-Net · NTU RGB+D (cross-subject) | AUC over observation ratios: 67.7 -> 70.6 | Improved |
| add real/fake loss to generator (ablation) | HARD-Net · NTU RGB+D (cross-subject) | AUC over observation ratios: 68.0 -> 70.6 | Improved |
| increase HI-IC bank size to 5000 and beyond (ablation) | HARD-Net · NTU RGB+D (cross-subject) | AUC over observation ratios | No clear effect |
| vary original-to-hard sample proportion from 1:1 to 6:1 (best 4:1) (ablation) | HARD-Net · NTU RGB+D (cross-subject) | AUC over observation ratios: 69.5 to 70.6 | No clear effect |
Later papers that compared with this method
3 later accepted papers report HardNet as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Descriptor Distillation: A Teacher-Student-Regularized Framework for Learning Local Descriptors (IJCV 2024 · better on 3 of 3 shared results)
- Revisiting Unsupervised Local Descriptor Learning (AAAI 2023 · better on 0 of 1 shared result)
- Online Invariance Selection for Local Feature Descriptors (ECCV 2020 · better on 4 of 4 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
- On Diverse Asynchronous Activity Anticipation (ECCV 2020 · similarity 0.83)
- Joint Prediction of Activity Labels and Starting Times in Untrimmed Videos (ICCV 2017 · similarity 0.82)
- Towards Realistic Predictors (ECCV 2018 · similarity 0.82)
- Learning Activity Progression in LSTMs for Activity Detection and Early Detection (CVPR 2016 · similarity 0.81)
- HARDVS: Revisiting Human Activity Recognition with Dynamic Vision Sensors (AAAI 2024 · similarity 0.80)
- Learning to Predict Activity Progress by Self-Supervised Video Alignment (CVPR 2024 · similarity 0.79)
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.
