Skip to content

Update README with current models and versioning docs #16

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
TurboGit merged 2 commits into from
Apr 12, 2026
Merged

Update README with current models and versioning docs #16

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
dff2d12
Update README with current models and versioning docs
andriiryzhkov Apr 5, 2026
34aa301
Remove DA2
andriiryzhkov Apr 8, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
39 changes: 30 additions & 9 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,19 +6,22 @@ Currently targets the ONNX backend. The pipeline is designed to support addition

## Models

| Model | Task | Description |
|---------------------------------------------------------------------------------|-------------|-----------------------------------------------|
| [`denoise-nafnet`](models/denoise-nafnet/README.md) | denoise | NAFNet denoiser trained on SIDD dataset |
| [`denoise-nind`](models/denoise-nind/README.md) | denoise | UNet denoiser trained on NIND dataset |
| [`mask-object-segnext-b2hq`](models/mask-object-segnext-b2hq/README.md) | mask-object | SegNext ViT-B SAx2 HQ for masking |
| [`upscale-bsrgan`](models/upscale-bsrgan/README.md) | upscale | BSRGAN 2x and 4x blind super-resolution |
| Model | Task | Description |
|-------------------------------------------------------------------------------|---------|-------------------------------------------------|
| [`denoise-nind`](models/denoise-nind/README.md) | denoise | UNet denoiser trained on NIND dataset |
| [`embed-openclip-vitb32`](models/embed-openclip-vitb32/README.md) | embed | OpenCLIP ViT-B/32 text/image embeddings |
| [`mask-object-sam21-base-plus`](models/mask-object-sam21-base-plus/README.md) | mask | SAM 2.1 Hiera Base Plus for interactive masking |
| [`mask-object-sam21-small`](models/mask-object-sam21-small/README.md) | mask | SAM 2.1 Hiera Small for interactive masking |
| [`mask-object-sam21-tiny`](models/mask-object-sam21-tiny/README.md) | mask | SAM 2.1 Hiera Tiny for interactive masking |
| [`mask-object-segnext-b2hq`](models/mask-object-segnext-b2hq/README.md) | mask | SegNext ViT-B SAx2 HQ for semantic masking |
| [`upscale-bsrgan`](models/upscale-bsrgan/README.md) | upscale | BSRGAN 2x and 4x blind super-resolution |

## Repository structure

```
pyproject.toml Project configuration, dependency groups, CLI entry point
darktable_ai/ Python package (CLI + pipeline orchestration)
vendor/ Git submodules (NAFNet, nind-denoise, SegNext)
vendor/ Git submodules (nind-denoise, sam2, SegNext)
samples/<task>/ Sample images organized by task
output/ Build output: ONNX models + config.json (gitignored)
temp/ Downloaded checkpoints (gitignored)
Expand All @@ -34,7 +37,7 @@ models/

Requires [uv](https://docs.astral.sh/uv/) and Python 3.11–3.12.

Dependencies are managed through [dependency groups](https://docs.astral.sh/uv/concepts/dependency-groups/) in `pyproject.toml`. The base package only needs `click` and `pyyaml`. ML dependencies are split into groups — one per model plus a shared `core` group — so you only install what you need. Use `uv sync --group <name>` to install a specific group, or `--group all-models` for everything.
Dependencies are managed through [dependency groups](https://docs.astral.sh/uv/concepts/dependency-groups/) in `pyproject.toml`. The base package only needs `click` and `pyyaml`. ML dependencies are split into groups – one per model plus a shared `core` group – so you only install what you need. Use `uv sync --group <name>` to install a specific group, or `--group all-models` for everything.

## Setup

Expand Down Expand Up @@ -71,18 +74,36 @@ uv run dtai validate denoise-nind # Validate ONNX output
uv run dtai package denoise-nind # Create .dtmodel archive
uv run dtai demo denoise-nind # Run demo on sample images

# Generate versions.json
uv run dtai versions

# Evaluate a model
uv sync --group eval
uv run dtai eval mask mask-object-segnext-b2hq --limit 5
```

## Versioning

Each model has a `version` field in its `model.yaml` that gets written to `config.json`. The `dtai versions` command generates `output/versions.json` – a manifest mapping model IDs to their current versions:

```json
{
"models": {
"denoise-nind": "1.0",
"mask-object-sam21-small": "1.0"
}
}
```

This file is published as a GitHub release asset alongside `.dtmodel` packages. darktable fetches it to check for model updates without downloading full packages.

## Demos

Each model includes a `demo.py` script that runs inference on sample images
from `samples/<task>/`. Models that require per-image input (e.g. point prompts
for object segmentation) define `image_args` in their `model.yaml`.

Output images are saved to `models/<model>/output/`.
Output images are saved to `output/<model>-demo/`.

## Model selection criteria

Expand Down
Loading