The official code of "Deep Regression Representation Learning with Topology" in ICML 2024. [PDF]. [Project Page].
Experiments on the synthetic dataset
- main.py : Obtain experimental results on the synthetic dataset
- visualization.py: Visualizing the feature space, the input space and the target space.
- PH_Reg.py: Code of our proposed PH-Reg.
- ordinal_entropy.py: Code of the regression baseline
You can generate your own synthetic dataset by:
- run data/synthetze_datasets.py
Experiments on Real-world Datasets
We currently provide a detailed code for age estimation. PH-reg can be added in a similar way for other tasks.
For age estimation, we follow the Imbalanced Regression, please download Age-DB dataset following their instructions.
Acknowledgement
Ph-Reg is built based on the topological autoencoder and Birdal's regularizer.
An old version of the paper can be found here. Shihao wanted to express his gratitude for the high-quality comments made by the reviewers, which definitely improved the paper.