tdpeuter/2025ML-project-neural_compression
tdpeuter/2025ML-project-neural_compression
Archived
1
Fork 0
Code Issues Pull requests Releases 1 Activity Actions

feat: initial for IDF

Browse source
This commit is contained in:
Robin Meersman 2025-11-07 12:54:36 +01:00
commit ef4684ef39
27 changed files with 2830 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

29
integer_discrete_flows/README.md Normal file
View file

@ -0,0 +1,29 @@
# Integer Discrete Flows and Lossless Compression
This repository contains the code for the experiments presented in [1].
## Usage
### CIFAR10 setup:
```
python main_experiment.py --n_flows 8 --n_levels 3 --n_channels 512 --coupling_type 'densenet' --densenet_depth 12 —n_mixtures 5 —splitprior_type ‘densenet’
```
### ImageNet32 setup:
```
python main_experiment.py --evaluate_interval_epochs 5 --n_flows 8 --n_levels 3 --n_channels 512 --n_mixtures 5 --densenet_depth 12 --coupling_type 'densenet' --splitprior_type 'densenet' --dataset 'imagenet32' --epochs 100 --lr_decay 0.99
```
### ImageNet64 setup:
```
python main_experiment.py --evaluate_interval_epochs 1 --n_flows 8 --n_levels 4 --n_channels 512 --n_mixtures 5 --densenet_depth 12 --coupling_type 'densenet' --splitprior_type 'densenet' --dataset 'imagenet64' --epochs 20 --lr_decay 0.99 --batch_size 64
```
# Acknowledgements
The Robert Bosch GmbH is acknowledged for financial support.
# References
[1] Hoogeboom, Emiel, Jorn WT Peters, Rianne van den Berg, and Max Welling. "Integer Discrete Flows and Lossless Compression." Conference on Neural Information Processing Systems (2019).