Since earlier this year, I’ve been developing a machine-learning library called medicai, built entirely on Keras 3 with full multi-backend support (tensorflow, torch, jax). It handles both 2D and 3D inputs. Although the project is still new and actively evolving, the models available in the main branch are already stable and ready to use. You can try them for 2D, 2.5D, and 3D tasks. Notable models include:
Repository
\nTutorials:
\nSince earlier this year, I’ve been developing a machine-learning library called medicai, built entirely on Keras 3 with full multi-backend support (tensorflow, torch, jax). It handles both 2D and 3D inputs. Although the project is still new and actively evolving, the models available in the main branch are already stable and ready to use. You can try them for 2D, 2.5D, and 3D tasks. Notable models include:
Repository
\nTutorials:
\namazing work how do you train these ?
", "rawMarkdown": "amazing work how do you train these ?", "votes": null }, { "id": "3351121", "postDate": "11/28/2025 06:57:07", "content": "You can train any model with your preferred library (tf/torch/jax) by setting the keras backend. The tutorials mentioned above are some starter. Here are more of it.
", "rawMarkdown": "You can train any model with your preferred library (tf/torch/jax) by setting the keras backend. The tutorials mentioned above are some starter. [Here](https://github.com/innat/medic-ai?tab=readme-ov-file#-guides) are more of it.", "votes": null } ], "comments": [ { "id": 3351014, "author_name": "theonetheonly", "author_url": "", "post_date": "11/28/2025 04:52:58", "content": "amazing work how do you train these ?
", "votes": null, "replies": [ { "id": 3351121, "author_name": "ipythonx", "author_url": "", "post_date": "11/28/2025 06:57:07", "content": "You can train any model with your preferred library (tf/torch/jax) by setting the keras backend. The tutorials mentioned above are some starter. Here are more of it.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "3328581": "Since earlier this year, I’ve been developing a machine-learning library called `medicai`, built entirely on **Keras 3** with full multi-backend support (`tensorflow`, `torch`, `jax`). It handles both 2D and 3D inputs. Although the project is still new and actively evolving, the models available in the main branch are already stable and ready to use. You can try them for 2D, 2.5D, and 3D tasks. Notable models include:\n\n- [UNETR](https://github.com/innat/medic-ai/blob/main/medicai/models/unetr/README.md)\n- [UNETR++](https://github.com/innat/medic-ai/blob/main/medicai/models/unetr_plus_plus/README.md)\n- [SwinUNETR](https://github.com/innat/medic-ai/blob/main/medicai/models/swin/README.md)\n- [TransUNet](https://github.com/innat/medic-ai/blob/main/medicai/models/transunet/README.md)\n- [SegFormer](https://github.com/innat/medic-ai/blob/main/medicai/models/segformer/README.md)\n- [UPerNet](https://github.com/innat/medic-ai/blob/main/medicai/models/upernet/README.md)\n\n\n**Repository**\n\n- [medicai](https://github.com/innat/medic-ai)\n\n**Tutorials:**\n\n- [3D segmentation cases with Keras built-in training API with JAX backend - TPU](https://www.kaggle.com/code/ipythonx/vesuvius-surface-detection) - (use this comp dataset)\n- [3D segmentation cases with pure PyTorch + TFRecord](https://www.kaggle.com/code/ipythonx/vesuvius-surface-detection-in-pytorch) - (use this comp dataset)\n- [2.5D segmentation cases with Jax](https://www.kaggle.com/code/ipythonx/vesuvius-surface-2-5d-detection-in-jax) - (use this comp dataset)\n- [3D segmentation cases with PyTorch-lightning](https://www.kaggle.com/code/ipythonx/train-vesuvius-surface-3d-detection-in-lightning) - (use this comp dataset)\n- [3D Inference](https://www.kaggle.com/code/ipythonx/inference-vesuvius-surface-3d-detection) - (use this comp dataset)\n- [tif to tfrecord conversion](https://www.kaggle.com/code/ipythonx/vesuvius-tif-to-tfrecord)", "3351014": "amazing work how do you train these ?", "3351121": "You can train any model with your preferred library (tf/torch/jax) by setting the keras backend. The tutorials mentioned above are some starter. [Here](https://github.com/innat/medic-ai?tab=readme-ov-file#-guides) are more of it." }, "source": "meta" }