GarNet: A Two-stream Network for Fast and Accurate 3D Cloth Draping
Supplementary material
Paper ID 701
Please open this webpage in one of the following browsers, which support embedded videos:
  • Google Chrome >= 65.0.3325
  • Mozilla Firefox >= 59.0.1
  • Apple Safari >= 11.0.3

Overview

We first supply details about our garment-fitting dataset and the trained network models. We then provide videos depicting our results for each garment. Finally, we demonstrate the qualitative effectiveness of our penetration term.

Dataset and architectures

In the supplementary pdf document, you will find the details about our dataset and architecture along with qualitative results on the dataset of Wang et al. 2018. Finally, the interpenetration term of our loss function is further explained:

Please click here for supplementary pdf.

Result Videos

1. Prediction Results in our dataset

The following videos show our predictions and the PBS side-by-side for different bodies and different poses.

Jeans

Sweater

T-shirt

2. Architecture Comparison

The following video shows the GarNet-Local, GarNet-Global, GarNet-Naive and PBS results side-by-side.

3. Ablation Study

The following video depicts our results when turning off the penetration term in the loss function during training.