Research map / 3D vision / Face and shape
CNN-Based Real-Time Dense Face Reconstruction with Inverse-Rendered Photo-Realistic Face Images
Yudong Guo, Juyong Zhang, Jianfei Cai, Boyi Jiang, Jianmin Zheng
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 FineNet on top of Tracking CoarseNet (ablation) | Tracking CoarseNet · FaceCap (200 frames, video) | average point-to-point distance (mm): 2.11 -> 2.08 | Improved |
| add FineNet to CoarseNet on a wrinkle-rich subregion (ablation) | CoarseNet · Wrinkle-rich subregion of a video frame | mean error (mm): 2.20 -> 2.03 | Improved |
| use proposed per-pixel projection loss instead of prior CoarseNet loss (ablation) | Single-image CoarseNet · AFLW test (~700 images, CoarseNet 15000 iterations) | L_pose test error: 26.35 -> 7.69 | Improved |
| use proposed per-pixel projection loss instead of prior CoarseNet loss (ablation) | Single-image CoarseNet · AFLW test (~700 images, CoarseNet 15000 iterations) | L_geo test error: 5.53 -> 4.23 | Improved |
| use proposed per-pixel projection loss instead of prior CoarseNet loss (ablation) | Single-image CoarseNet · AFLW test (~700 images, CoarseNet 15000 iterations) | MSE (pose parameters): 1.91 -> 0.56 | Improved |
| use proposed per-pixel projection loss instead of prior CoarseNet loss (ablation) | Single-image CoarseNet · AFLW test (~700 images, CoarseNet 15000 iterations) | mean vertex distance (geometry parameters): 5.18 -> 4.55 | Improved |
| use learned CoarseNet+FineNet instead of optimization-based inverse rendering | face reconstruction from video · FaceCap (200 frames, video) | average point-to-point distance (mm): 1.81 -> 2.08 | Mixed |
| use proposed method instead of the strongest learning-based single-image baseline | Single-image CoarseNet+FineNet · FRGC v2 Spring2004range (2100 images) | RMSE (mm): 5.367 -> 4.915 | Improved |
| use proposed method instead of the strongest learning-based single-image baseline | Single-image CoarseNet+FineNet · FRGC v2 Spring2004range (2100 images) | MAE (mm): 3.923 -> 3.846 | Improved |
| use proposed learning-based tracker instead of optimization-based tracker | dense face tracking · Video frame with rich details (binocular ground truth) | mean error (mm): 1.96 -> 1.8 | No clear effect |
| use Tracking CoarseNet with previous-frame pose instead of Single-image CoarseNet on each frame (ablation) | video face tracking · Video tracking (qualitative) | tracking robustness | Improved |
| use learned networks instead of optimization-based inverse rendering for single images | single-image reconstruction · Large pose and strong lighting images (qualitative) | robustness | Improved |
Closest papers by meaning
- Learning Detailed Face Reconstruction From a Single Image (CVPR 2017 · similarity 0.93)
- High-Fidelity Monocular Face Reconstruction Based on an Unsupervised Model-Based Face Autoencoder (TPAMI 2018 · similarity 0.90)
- InverseFaceNet: Deep Monocular Inverse Face Rendering (CVPR 2018 · similarity 0.90)
- Photorealistic Facial Texture Inference Using Deep Neural Networks (CVPR 2017 · similarity 0.90)
- Towards High-Fidelity 3D Face Reconstruction From In-the-Wild Images Using Graph Convolutional Networks (CVPR 2020 · similarity 0.90)
- Regressing Robust and Discriminative 3D Morphable Models With a Very Deep Neural Network (CVPR 2017 · similarity 0.89)
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.
