cv-suite 0.0.1__tar.gz

cv_suite-0.0.1/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Pedro Tonso
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

cv_suite-0.0.1/PKG-INFO

@@ -0,0 +1,115 @@
+Metadata-Version: 2.4
+Name: cv-suite
+Version: 0.0.1
+Summary: Unified computer vision toolkit with shared FM runtime and branch-owned I/O.
+Author-email: Pedro Tonso <ptonso@gmail.com>
+License-Expression: MIT
+Project-URL: Homepage, https://github.com/ptonso/cv-suite
+Project-URL: Repository, https://github.com/ptonso/cv-suite
+Project-URL: Issues, https://github.com/ptonso/cv-suite/issues
+Keywords: computer-vision,dataset,annotation,foundation-models,cli,yolo,coco
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: Science/Research
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering :: Image Processing
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: numpy>=1.23
+Requires-Dist: pillow>=9.5
+Requires-Dist: imagehash>=4.3
+Requires-Dist: opencv-python>=4.8
+Requires-Dist: pyyaml>=6.0
+Requires-Dist: tqdm>=4.66
+Provides-Extra: dev
+Requires-Dist: pytest>=7.4; extra == "dev"
+Requires-Dist: pytest-cov>=4.1; extra == "dev"
+Requires-Dist: ruff>=0.6; extra == "dev"
+Requires-Dist: mypy>=1.7; extra == "dev"
+Dynamic: license-file
+
+# cvsuite
+
+`cvsuite` is a computer vision toolkit for reproducible, filesystem-to-filesystem dataset work, organized around one CLI:
+
+```bash
+cvsuite <branch> <src> [transform] <output>
+```
+
+Convert between annotation formats (YOLO, COCO, LabelMe, class folders, segmentation masks, VQA manifests). Auto-label raw images from a text prompt with foundation grounding models such as SAM 3 or Grounding DINO. Caption them or answer questions about them with vision-language models. Sort them into class folders with zero-shot classifiers. Read text off them with OCR. Generate and edit images with diffusion models. Then do the unglamorous half of the job: unzip and flatten a dump of files, drop exact and near-duplicates, fix orientation, normalize size and format, and draw a balanced sample for training.
+
+```bash
+cvsuite label ./dataset-yolo to-coco ./dataset-coco
+cvsuite label ./images ground --provider gsam --prompt "fire, smoke" to-coco ./labeled
+cvsuite vlm ./images caption --provider qwen to-json ./captions
+cvsuite class ./images infer --provider clip --prompt "good,bad" to-class-dir ./sorted
+cvsuite gen create --provider flux --prompt "a red fox in the snow" to-dst ./fox.png
+cvsuite prep arrange ./dump ./clean --unzip --exact-dedup --rename-seq
+```
+
+More in the [cookbook](https://github.com/ptonso/cv-suite/blob/main/docs/cookbook.md).
+
+You point a branch at a source, optionally run a transform, and materialize a destination layout. There is deliberately no universal "convert anything to anything" command.
+
+## How it works
+
+```mermaid
+flowchart LR
+    SRC["source<br/>yolo · coco · labelme<br/>images · shards · …"] --> TOOL
+
+    TOOL{"pick a tool<br/>label · class · vlm · gen"} -->|ingest| DS["VisionDataset"]
+
+    DS --> TF["one transform<br/>sample · ground · ocr<br/>infer · caption · vqa"]
+    TF -->|"chain runs to stack"| DS
+
+    DS -->|output command| DST["destination<br/>yolo · coco · class-dir<br/>json · shards · …"]
+```
+
+Whichever tool you pick, the spine is the same: the source is ingested into a shared `VisionDataset`, transforms mutate that dataset, and an output command materializes it. What differs per tool is the ingest heuristics, which transforms exist, and the export rules.
+
+Each run applies at most one transform, so you stack them by chaining runs: the destination of one becomes the source of the next. FM-backed transforms (`ground`, `ocr`, `infer`, `caption`, `vqa`, and the `gen` actions) hand the dataset to an isolated provider subprocess that mutates records in place. `cvsuite prep` is the exception to all of this, since it works directly on folders and never builds a dataset.
+
+## Branches
+
+| Branch | Purpose |
+|---|---|
+| `cvsuite label` | annotated datasets: YOLO / COCO / LabelMe / image-only ingest, `sample` / `ground` / `ocr` / `ops` / `filter` transforms, export to any label format |
+| `cvsuite class` | class-directory datasets: `infer` with zero-shot classifiers, class-aware `sample`, export via `to-class-dir` |
+| `cvsuite vlm` | vision-language datasets: `caption` / `ask` / `vqa` / `map-answers`, export to flat JSON / shards / VQA-style |
+| `cvsuite gen` | text-to-image and image-edit generation, materialized via `to-dst` |
+| `cvsuite prep` | raw image folder cleanup: `arrange` / `process` / `sample` / `orient`, file-first |
+| `cvsuite common` | shared operations: FM cache paths, inspection, and purge |
+
+## Install
+
+```bash
+pip install cv-suite
+```
+
+Or from source:
+
+```bash
+git clone https://github.com/ptonso/cv-suite
+pip install -e ./cv-suite
+```
+
+The base install ships the CLI and the I/O layer. Foundation-model backends run in isolated per-model virtualenvs provisioned on first use. See [`docs/getting-started.md`](https://github.com/ptonso/cv-suite/blob/main/docs/getting-started.md).
+
+## Documentation
+
+- [`docs/`](https://github.com/ptonso/cv-suite/tree/main/docs) contains the guides: the [mental model](https://github.com/ptonso/cv-suite/blob/main/docs/index.md), [getting started](https://github.com/ptonso/cv-suite/blob/main/docs/getting-started.md), and a [cookbook](https://github.com/ptonso/cv-suite/blob/main/docs/cookbook.md) of common workflows.
+- [`specs/`](https://github.com/ptonso/cv-suite/tree/main/specs) is the technical source of truth: architecture, the dataset contract, per-branch behavior, business decisions, and the guardrails that must not break.
+
+## Names
+
+- PyPI distribution: `cv-suite`
+- import package and CLI: `cvsuite`
+
+## License
+
+MIT. See [LICENSE](https://github.com/ptonso/cv-suite/blob/main/LICENSE).

cv_suite-0.0.1/README.md

@@ -0,0 +1,80 @@
+# cvsuite
+
+`cvsuite` is a computer vision toolkit for reproducible, filesystem-to-filesystem dataset work, organized around one CLI:
+
+```bash
+cvsuite <branch> <src> [transform] <output>
+```
+
+Convert between annotation formats (YOLO, COCO, LabelMe, class folders, segmentation masks, VQA manifests). Auto-label raw images from a text prompt with foundation grounding models such as SAM 3 or Grounding DINO. Caption them or answer questions about them with vision-language models. Sort them into class folders with zero-shot classifiers. Read text off them with OCR. Generate and edit images with diffusion models. Then do the unglamorous half of the job: unzip and flatten a dump of files, drop exact and near-duplicates, fix orientation, normalize size and format, and draw a balanced sample for training.
+
+```bash
+cvsuite label ./dataset-yolo to-coco ./dataset-coco
+cvsuite label ./images ground --provider gsam --prompt "fire, smoke" to-coco ./labeled
+cvsuite vlm ./images caption --provider qwen to-json ./captions
+cvsuite class ./images infer --provider clip --prompt "good,bad" to-class-dir ./sorted
+cvsuite gen create --provider flux --prompt "a red fox in the snow" to-dst ./fox.png
+cvsuite prep arrange ./dump ./clean --unzip --exact-dedup --rename-seq
+```
+
+More in the [cookbook](https://github.com/ptonso/cv-suite/blob/main/docs/cookbook.md).
+
+You point a branch at a source, optionally run a transform, and materialize a destination layout. There is deliberately no universal "convert anything to anything" command.
+
+## How it works
+
+```mermaid
+flowchart LR
+    SRC["source<br/>yolo · coco · labelme<br/>images · shards · …"] --> TOOL
+
+    TOOL{"pick a tool<br/>label · class · vlm · gen"} -->|ingest| DS["VisionDataset"]
+
+    DS --> TF["one transform<br/>sample · ground · ocr<br/>infer · caption · vqa"]
+    TF -->|"chain runs to stack"| DS
+
+    DS -->|output command| DST["destination<br/>yolo · coco · class-dir<br/>json · shards · …"]
+```
+
+Whichever tool you pick, the spine is the same: the source is ingested into a shared `VisionDataset`, transforms mutate that dataset, and an output command materializes it. What differs per tool is the ingest heuristics, which transforms exist, and the export rules.
+
+Each run applies at most one transform, so you stack them by chaining runs: the destination of one becomes the source of the next. FM-backed transforms (`ground`, `ocr`, `infer`, `caption`, `vqa`, and the `gen` actions) hand the dataset to an isolated provider subprocess that mutates records in place. `cvsuite prep` is the exception to all of this, since it works directly on folders and never builds a dataset.
+
+## Branches
+
+| Branch | Purpose |
+|---|---|
+| `cvsuite label` | annotated datasets: YOLO / COCO / LabelMe / image-only ingest, `sample` / `ground` / `ocr` / `ops` / `filter` transforms, export to any label format |
+| `cvsuite class` | class-directory datasets: `infer` with zero-shot classifiers, class-aware `sample`, export via `to-class-dir` |
+| `cvsuite vlm` | vision-language datasets: `caption` / `ask` / `vqa` / `map-answers`, export to flat JSON / shards / VQA-style |
+| `cvsuite gen` | text-to-image and image-edit generation, materialized via `to-dst` |
+| `cvsuite prep` | raw image folder cleanup: `arrange` / `process` / `sample` / `orient`, file-first |
+| `cvsuite common` | shared operations: FM cache paths, inspection, and purge |
+
+## Install
+
+```bash
+pip install cv-suite
+```
+
+Or from source:
+
+```bash
+git clone https://github.com/ptonso/cv-suite
+pip install -e ./cv-suite
+```
+
+The base install ships the CLI and the I/O layer. Foundation-model backends run in isolated per-model virtualenvs provisioned on first use. See [`docs/getting-started.md`](https://github.com/ptonso/cv-suite/blob/main/docs/getting-started.md).
+
+## Documentation
+
+- [`docs/`](https://github.com/ptonso/cv-suite/tree/main/docs) contains the guides: the [mental model](https://github.com/ptonso/cv-suite/blob/main/docs/index.md), [getting started](https://github.com/ptonso/cv-suite/blob/main/docs/getting-started.md), and a [cookbook](https://github.com/ptonso/cv-suite/blob/main/docs/cookbook.md) of common workflows.
+- [`specs/`](https://github.com/ptonso/cv-suite/tree/main/specs) is the technical source of truth: architecture, the dataset contract, per-branch behavior, business decisions, and the guardrails that must not break.
+
+## Names
+
+- PyPI distribution: `cv-suite`
+- import package and CLI: `cvsuite`
+
+## License
+
+MIT. See [LICENSE](https://github.com/ptonso/cv-suite/blob/main/LICENSE).

cv_suite-0.0.1/pyproject.toml

@@ -0,0 +1,61 @@
+[build-system]
+requires = ["setuptools>=77", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "cv-suite"
+version = "0.0.1"
+description = "Unified computer vision toolkit with shared FM runtime and branch-owned I/O."
+readme = "README.md"
+requires-python = ">=3.10"
+license = "MIT"
+license-files = ["LICENSE"]
+authors = [{ name = "Pedro Tonso", email = "ptonso@gmail.com" }]
+keywords = ["computer-vision", "dataset", "annotation", "foundation-models", "cli", "yolo", "coco"]
+classifiers = [
+  "Development Status :: 3 - Alpha",
+  "Intended Audience :: Developers",
+  "Intended Audience :: Science/Research",
+  "Programming Language :: Python :: 3",
+  "Programming Language :: Python :: 3.10",
+  "Programming Language :: Python :: 3.11",
+  "Programming Language :: Python :: 3.12",
+  "Topic :: Scientific/Engineering :: Image Processing",
+  "Topic :: Scientific/Engineering :: Artificial Intelligence",
+]
+dependencies = [
+  "numpy>=1.23",
+  "pillow>=9.5",
+  "imagehash>=4.3",
+  "opencv-python>=4.8",
+  "pyyaml>=6.0",
+  "tqdm>=4.66",
+]
+
+[project.optional-dependencies]
+dev = [
+  "pytest>=7.4",
+  "pytest-cov>=4.1",
+  "ruff>=0.6",
+  "mypy>=1.7",
+]
+
+[project.urls]
+Homepage = "https://github.com/ptonso/cv-suite"
+Repository = "https://github.com/ptonso/cv-suite"
+Issues = "https://github.com/ptonso/cv-suite/issues"
+
+[project.scripts]
+cvsuite = "cvsuite.cli:main"
+
+[tool.setuptools.packages.find]
+where = ["src"]
+include = ["cvsuite*"]
+
+[tool.setuptools.package-data]
+cvsuite = ["common/fm/providers/setup_venv/*.sh"]
+
+[tool.pytest.ini_options]
+minversion = "7.0"
+addopts = "-q"
+pythonpath = ["src"]

cv_suite-0.0.1/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

cv_suite-0.0.1/src/cv_suite.egg-info/PKG-INFO

@@ -0,0 +1,115 @@
+Metadata-Version: 2.4
+Name: cv-suite
+Version: 0.0.1
+Summary: Unified computer vision toolkit with shared FM runtime and branch-owned I/O.
+Author-email: Pedro Tonso <ptonso@gmail.com>
+License-Expression: MIT
+Project-URL: Homepage, https://github.com/ptonso/cv-suite
+Project-URL: Repository, https://github.com/ptonso/cv-suite
+Project-URL: Issues, https://github.com/ptonso/cv-suite/issues
+Keywords: computer-vision,dataset,annotation,foundation-models,cli,yolo,coco
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: Science/Research
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering :: Image Processing
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: numpy>=1.23
+Requires-Dist: pillow>=9.5
+Requires-Dist: imagehash>=4.3
+Requires-Dist: opencv-python>=4.8
+Requires-Dist: pyyaml>=6.0
+Requires-Dist: tqdm>=4.66
+Provides-Extra: dev
+Requires-Dist: pytest>=7.4; extra == "dev"
+Requires-Dist: pytest-cov>=4.1; extra == "dev"
+Requires-Dist: ruff>=0.6; extra == "dev"
+Requires-Dist: mypy>=1.7; extra == "dev"
+Dynamic: license-file
+
+# cvsuite
+
+`cvsuite` is a computer vision toolkit for reproducible, filesystem-to-filesystem dataset work, organized around one CLI:
+
+```bash
+cvsuite <branch> <src> [transform] <output>
+```
+
+Convert between annotation formats (YOLO, COCO, LabelMe, class folders, segmentation masks, VQA manifests). Auto-label raw images from a text prompt with foundation grounding models such as SAM 3 or Grounding DINO. Caption them or answer questions about them with vision-language models. Sort them into class folders with zero-shot classifiers. Read text off them with OCR. Generate and edit images with diffusion models. Then do the unglamorous half of the job: unzip and flatten a dump of files, drop exact and near-duplicates, fix orientation, normalize size and format, and draw a balanced sample for training.
+
+```bash
+cvsuite label ./dataset-yolo to-coco ./dataset-coco
+cvsuite label ./images ground --provider gsam --prompt "fire, smoke" to-coco ./labeled
+cvsuite vlm ./images caption --provider qwen to-json ./captions
+cvsuite class ./images infer --provider clip --prompt "good,bad" to-class-dir ./sorted
+cvsuite gen create --provider flux --prompt "a red fox in the snow" to-dst ./fox.png
+cvsuite prep arrange ./dump ./clean --unzip --exact-dedup --rename-seq
+```
+
+More in the [cookbook](https://github.com/ptonso/cv-suite/blob/main/docs/cookbook.md).
+
+You point a branch at a source, optionally run a transform, and materialize a destination layout. There is deliberately no universal "convert anything to anything" command.
+
+## How it works
+
+```mermaid
+flowchart LR
+    SRC["source<br/>yolo · coco · labelme<br/>images · shards · …"] --> TOOL
+
+    TOOL{"pick a tool<br/>label · class · vlm · gen"} -->|ingest| DS["VisionDataset"]
+
+    DS --> TF["one transform<br/>sample · ground · ocr<br/>infer · caption · vqa"]
+    TF -->|"chain runs to stack"| DS
+
+    DS -->|output command| DST["destination<br/>yolo · coco · class-dir<br/>json · shards · …"]
+```
+
+Whichever tool you pick, the spine is the same: the source is ingested into a shared `VisionDataset`, transforms mutate that dataset, and an output command materializes it. What differs per tool is the ingest heuristics, which transforms exist, and the export rules.
+
+Each run applies at most one transform, so you stack them by chaining runs: the destination of one becomes the source of the next. FM-backed transforms (`ground`, `ocr`, `infer`, `caption`, `vqa`, and the `gen` actions) hand the dataset to an isolated provider subprocess that mutates records in place. `cvsuite prep` is the exception to all of this, since it works directly on folders and never builds a dataset.
+
+## Branches
+
+| Branch | Purpose |
+|---|---|
+| `cvsuite label` | annotated datasets: YOLO / COCO / LabelMe / image-only ingest, `sample` / `ground` / `ocr` / `ops` / `filter` transforms, export to any label format |
+| `cvsuite class` | class-directory datasets: `infer` with zero-shot classifiers, class-aware `sample`, export via `to-class-dir` |
+| `cvsuite vlm` | vision-language datasets: `caption` / `ask` / `vqa` / `map-answers`, export to flat JSON / shards / VQA-style |
+| `cvsuite gen` | text-to-image and image-edit generation, materialized via `to-dst` |
+| `cvsuite prep` | raw image folder cleanup: `arrange` / `process` / `sample` / `orient`, file-first |
+| `cvsuite common` | shared operations: FM cache paths, inspection, and purge |
+
+## Install
+
+```bash
+pip install cv-suite
+```
+
+Or from source:
+
+```bash
+git clone https://github.com/ptonso/cv-suite
+pip install -e ./cv-suite
+```
+
+The base install ships the CLI and the I/O layer. Foundation-model backends run in isolated per-model virtualenvs provisioned on first use. See [`docs/getting-started.md`](https://github.com/ptonso/cv-suite/blob/main/docs/getting-started.md).
+
+## Documentation
+
+- [`docs/`](https://github.com/ptonso/cv-suite/tree/main/docs) contains the guides: the [mental model](https://github.com/ptonso/cv-suite/blob/main/docs/index.md), [getting started](https://github.com/ptonso/cv-suite/blob/main/docs/getting-started.md), and a [cookbook](https://github.com/ptonso/cv-suite/blob/main/docs/cookbook.md) of common workflows.
+- [`specs/`](https://github.com/ptonso/cv-suite/tree/main/specs) is the technical source of truth: architecture, the dataset contract, per-branch behavior, business decisions, and the guardrails that must not break.
+
+## Names
+
+- PyPI distribution: `cv-suite`
+- import package and CLI: `cvsuite`
+
+## License
+
+MIT. See [LICENSE](https://github.com/ptonso/cv-suite/blob/main/LICENSE).