带有注意力的液体扭曲 GAN: 人脸图像合成的统一框架
15/04/2021, iPERCore-0.2.0, including
Add Training
Add A Unified synthesizer: Motion Imitation + Appearance Transfer + Novel View Synthesis
Update torch 1.8+ with RTX30+ GPUs.
See the details of developing logs.
Liquid Warping GAN with Attention: A Unified Framework for Human Image Synthesis, including human motion imitation, appearance transfer, and novel view synthesis. Currently the paper is under review of IEEE TPAMI. It is an extension of our previous ICCV project impersonator, and it has a more powerful ability in generalization and produces higher-resolution results (512 x 512, 1024 x 1024) than the previous ICCV version.
| Colab Notebook | Paper | Website | Dataset | Bilibili |
|---|---|---|---|---|
| paper | website | Dataset | bilibili |
See more details, including system dependencies, python requirements and setups in install.md. Please follows the instructions in install.md to install this firstly.
Notice that imags_size=512 need at least 9.8GB GPU memory. if you are using a middle-level GPU(e.g. RTX 2060), you should change the image_size to 384 or 256. The following table can be used as a reference:
| image_size | preprocess | personalize | run_imitator | recommended gpu |
|---|---|---|---|---|
| 256x256 | 3.1 GB | 4.3 GB | 1.1 GB | RTX 2060 / RTX 2070 |
| 384x384 | 3.1 GB | 7.9 GB | 1.5 GB | GTX 1080Ti / RTX 2080Ti / Titan Xp |
| 512x512 | 3.1 GB | 9.8 GB | 2 GB | GTX 1080Ti / RTX 2080Ti / Titan Xp |
| 1024x1024 | 3.1 GB | 20 GB | - | RTX Titan / P40 / V100 32G |
See scripts_runner for more details.
…暂无开放 Issues，或尚未同步最近议题。