Code for paper "Cross-Domain Self-supervised Multi-task Feature Learning using Synthetic Imagery", Ren et al., CVPR'18
If you feel this useful, please consider cite:
@inproceedings{ren-cvpr2018,
title = {Cross-Domain Self-supervised Multi-task Feature Learning using Synthetic Imagery},
author = {Ren, Zhongzheng and Lee, Yong Jae},
booktitle = {IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
year = {2018}
}
Feel free to contact Jason Ren ([email protected]) if you have any questions!
git clone https://github.com/jason718/game-feature-learning
cd game-feature-learning
Since I greatly changed the code structure, I am retraining using the new code to reproduce the paper results.
SUNCG: Download the SUNCG images from suncg website. And make sure to put the files as the following structure:
suncg_image
├── depth
├── room_id1
├── ...
├── normal
├── room_id1
├── ...
├── edge
├── room_id1
├── ...
└── lab
├── room_id1
├── ...
SceneNet: Download the SceneNet images from scenenet website. And make sure to put the files as the following structure:
scenenet_image
└── train
├── 1
├── 2
├── ...
Please check scripts/surface_normals_code to generate surface normals from depth maps.
Dataset For Domain Adaptation:
sh ./scripts/train.sh
Evaluate on feature learning
Evaluate on three tasks
There are lots of awesome papers studying self-supervision for various tasks such as Image/Video Representation learning, Reinforcement learning, and Robotics. I am maintaining a paper list [awesome-self-supervised-learning] on Github. You are more than welcome to contribute and share :)
Supervised Learning is awesome but limited. Un-/Self-supervised learning generalizes better and sometimes also works better (which is already true in some geometry tasks)!
This work was supported in part by the National Science Foundation under Grant No. 1748387, the AWS Cloud Credits for Research Program, and GPUs donated by NVIDIA. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.