Research map / Adversarial robustness and security / Attacks on text and graphs
A+D Net: Training a Shadow Detector with Adversarial Shadow Attenuation
Hieu Le, Tomas F. Yago Vicente, Vu Nguyen, Minh Hoai, Dimitris Samaras
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 adversarially attenuated training images from A-Net (ablation) | D-Net · SBU test | BER: 8.8 -> 5.4 | Improved |
| add adversarially attenuated training images from A-Net (ablation) | D-Net · UCF test (trained on SBU) | BER: 11.8 -> 9.4 | Improved |
| add physics-based loss L_ph to A-Net (ablation) | D-Net without lambda_adv · SBU test | BER: 7.1 -> 5.7 | Improved |
| add physics-based loss L_ph to A-Net (ablation) | D-Net without lambda_adv · UCF test (trained on SBU) | BER: 13.6 -> 9.9 | Improved |
| add shadow-strength-based weight lambda_adv for adversarial examples (ablation) | D-Net with L_ph · SBU test | BER: 5.7 -> 5.4 | Improved |
| add shadow-strength-based weight lambda_adv for adversarial examples (ablation) | D-Net with L_ph · UCF test (trained on SBU) | BER: 9.9 -> 9.4 | Improved |
| increase adversarial loss weight lambda_sd from 1 to 10 (ablation) | A-Net / D-Net · SBU test | BER: 5.4 -> 6.5 | Hurt |
| rebalance real and adversarial weights to lambda_adv0=0.5 and lambda_real=0.5 (ablation) | D-Net · SBU test | BER: 5.4 -> 7.0 | Hurt |
| add D-Net with A-Net versus ST-CGAN (shadow error rate) | D-Net · SBU test | shadow error rate: 3.7 -> 5.3 | Hurt |
Closest papers by meaning
- ARGAN: Attentive Recurrent Generative Adversarial Network for Shadow Detection and Removal (ICCV 2019 · similarity 0.90)
- Stacked Conditional Generative Adversarial Networks for Jointly Learning Shadow Detection and Shadow Removal (CVPR 2018 · similarity 0.89)
- Shadow Detection With Conditional Generative Adversarial Networks (ICCV 2017 · similarity 0.89)
- From Shadow Segmentation to Shadow Removal (ECCV 2020 · similarity 0.88)
- Mask-ShadowGAN: Learning to Remove Shadows From Unpaired Data (ICCV 2019 · similarity 0.88)
- DC-ShadowNet: Single-Image Hard and Soft Shadow Removal Using Unsupervised Domain-Classifier Guided Network (ICCV 2021 · 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.
