Research map / Detection and segmentation / Open-set and few-shot detection
HLA-Face: Joint High-Low Adaptation for Low Light Face Detection
Wenjing Wang, Wenhan Yang, Jiaying Liu
Check 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 |
|---|---|---|---|
| Replace weak brightening with strong brightening (ablation) | HLA-Face · DARK FACE test | mAP (%): 38.3 -> 39.1 | Improved |
| Use jigsaw rather than rotation self-supervision alone (ablation) | HLA-Face · DARK FACE test | mAP (%): 22.7 -> 26.9 | Improved |
| Add rotation prediction to jigsaw self-supervision (ablation) | HLA-Face · DARK FACE test | mAP (%): 26.9 -> 25.3 | Hurt |
| Add cross-domain contrastive learning between H and D(H) (ablation) | Strong-brightening HLA-Face · DARK FACE test | mAP (%): 39.1 -> 40.9 | Improved |
| Use contrastive learning on H only instead of cross-domain contrastive learning (ablation) | Strong-brightening HLA-Face · DARK FACE test | mAP (%): 39.1 -> 38.2 | Hurt |
| Add pseudo labels on D(H) (ablation) | Strong-brightening HLA-Face · DARK FACE test | mAP (%): 39.1 -> 40.2 | Improved |
| Add jigsaw adaptation between E(L) and H (ablation) | Strong-brightening HLA-Face · DARK FACE test | mAP (%): 39.1 -> 40.2 | Improved |
| Add single-domain contrastive learning on E(L) (ablation) | Strong-brightening HLA-Face with cross-domain contrastive learning · DARK FACE test | mAP (%): 40.9 -> 41.1 | Improved |
| Combine jigsaw, cross-domain contrastive, and E(L) contrastive learning (ablation) | HLA-Face · DARK FACE test | mAP (%): 41.1 -> 41.4 | Improved |
| Add DSFD pyramid multi-scale testing (ablation) | HLA-Face · DARK FACE test | mAP (%): 41.4 -> 44.4 | Improved |
| Use HLA-Face without DARK FACE training labels instead of DSFD | DSFD · DARK FACE test | mAP (%): 16.1 -> 44.4 | Improved |
| Use CycleGAN-generated data with pseudo-label fine-tuning rather than CycleGAN-generated data alone | DSFD · DARK FACE test | mAP (%): 31.9% to 35.1% | 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.
- DSFD — DSFD: Dual Shot Face Detector (CVPR 2019) · better on 1 of 1 shared result
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
- Unsupervised Face Detection in the Dark (TPAMI 2022 · similarity 0.95)
- DLDA: Unified Dual-Level Domain Adaptation for Low-Light Object Detection (AAAI 2026 · similarity 0.85)
- Unsupervised Illumination Adaptation for Low-Light Vision (TPAMI 2024 · similarity 0.84)
- Boosting Object Detection with Zero-Shot Day-Night Domain Adaptation (CVPR 2024 · similarity 0.83)
- Low-Light Face Super-resolution via Illumination, Structure, and Texture Associated Representation (AAAI 2024 · similarity 0.83)
- You Only Look Around: Learning Illumination-Invariant Feature for Low-light Object Detection (NeurIPS 2024 · similarity 0.81)
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.
