| --- |
| datasets: |
| - kakaobrain/coyo-700m |
| language: |
| - en |
| --- |
| |
|
|
| ## Model Details |
|
|
| ### Model Description |
|
|
| This repo includes every models we trained during the Jax Community event sprint, organized by Hugging Face. |
| The folders {model} contains the Flax checkpoint and {model}_pt the Torch checkpoint. |
| |
| |
| |
| - **Developed by:** Baptiste Lemaire, Guillaume Thomas and Tom Dupuis from CEA-List |
| - **Model type:** Canny Edge Maps conditionned Diffusion model |
| - **Language(s) (NLP):** English |
| |
| |
| ## Uses |
| |
| <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. --> |
| - Fast low resolution image generation |
| - Online data augmentation |
| |
| See our gradio app for more information : [UCDR-Net gradio](https://huggingface.co/spaces/Baptlem/UCDR-Net) |
| |
| |
| ## Training Details |
| ### Training Data |
| |
| * [Coyo-700M](https://github.com/kakaobrain/coyo-dataset) |
| * [Bridge](https://sites.google.com/view/bridgedata) |
| |
| ### Training Procedure |
| We trained from scratch each one of our models. We kept the initial parameters, except for the Batch Size. |
| You can find the training script in the following [Event repo's folder](https://github.com/huggingface/community-events/blob/main/jax-controlnet-sprint/training_scripts/train_controlnet_flax.py) |
| |
| |
| #### Preprocessing |
| |
| -Resize to 128 resolution |
| -Canny Edge Map |
| |
| |
| #### Training parameters |
| The following table describes the differents hyperpa |
|  |
| |
| We stopped the coyo model a bit after it processed its first epoch. After running it, we discovered it performed pretty well even after only one epoch. So we deciced to keep it. |
| |
| |
| The last model has been trained with a custom DataLoader. The previous loads a batch containing 4 images from Bridge and 28 from Coyo. |
| Therefore, we can't talk about epoch as the model processed coyo faster than bridge. We then trained the model according to steps and not epoch. |
| |
| |
| |
| ### Results |
| |
| See [UCDR-Net gradio](https://huggingface.co/spaces/Baptlem/UCDR-Net) |
| |
| |
| |
| |
| ## Environmental Impact |
| |
| <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly --> |
| |
| Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). |
| |
| - **Hardware Type:** TPU v4 |
| - **Cloud Provider:** Gcloud |
| - **Compute Region:** us-central2-b |
| - **Carbon Emitted:** [More Information Needed] |
| |
| |
| |
| |
