smplfitter 0.2.10__tar.gz → 0.3.2.dev0__tar.gz

smplfitter-0.3.2.dev0/.editorconfig

@@ -0,0 +1,18 @@
+root = true
+
+[*]
+indent_style = space
+indent_size = 4
+end_of_line = lf
+charset = utf-8
+trim_trailing_whitespace = true
+insert_final_newline = true
+
+[*.{yml,yaml}]
+indent_size = 2
+
+[*.md]
+trim_trailing_whitespace = false
+
+[Makefile]
+indent_style = tab

smplfitter-0.3.2.dev0/.github/dependabot.yml

@@ -0,0 +1,11 @@
+version: 2
+updates:
+  - package-ecosystem: "pip"
+    directory: "/"
+    schedule:
+      interval: "weekly"
+
+  - package-ecosystem: "github-actions"
+    directory: "/"
+    schedule:
+      interval: "weekly"

smplfitter-0.3.2.dev0/.github/workflows/ci.yml

@@ -0,0 +1,27 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+jobs:
+  lint:
+    name: Lint
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.10"
+
+      - name: Install ruff
+        run: pip install ruff==0.4.10
+
+      - name: Lint with ruff
+        run: |
+          ruff check .
+          ruff format --check .

smplfitter-0.3.2.dev0/.github/workflows/python-publish.yml

@@ -0,0 +1,80 @@
+name: Build and Publish to PyPI
+
+on:
+  release:
+    types: [published]
+  workflow_dispatch:  # Allow manual trigger for testing
+
+permissions:
+  contents: read
+
+jobs:
+  lint:
+    name: Lint
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.10"
+
+      - name: Install ruff
+        run: pip install ruff==0.4.10
+
+      - name: Lint with ruff
+        run: |
+          ruff check .
+          ruff format --check .
+
+  build:
+    name: Build distribution
+    runs-on: ubuntu-latest
+    needs: lint
+
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0  # Required for setuptools-scm
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+
+      - name: Install build tools
+        run: python -m pip install --upgrade pip build
+
+      - name: Build wheel and sdist
+        run: python -m build
+
+      - name: Verify package
+        run: |
+          pip install dist/*.whl --no-deps
+          pip install numpy scipy chumpy --only-binary=:all: || pip install numpy scipy
+          python -c "import smplfitter; print(smplfitter.__version__)"
+
+      - uses: actions/upload-artifact@v4
+        with:
+          name: dist
+          path: dist/
+
+  publish:
+    name: Publish to PyPI
+    needs: build
+    runs-on: ubuntu-latest
+    environment: pypi
+    permissions:
+      id-token: write  # Required for trusted publishing
+    # Only publish on release, not on workflow_dispatch
+    if: github.event_name == 'release'
+
+    steps:
+      - uses: actions/download-artifact@v4
+        with:
+          name: dist
+          path: dist/
+
+      - name: Publish to PyPI
+        uses: pypa/gh-action-pypi-publish@release/v1

smplfitter-0.3.2.dev0/.pre-commit-config.yaml

@@ -0,0 +1,17 @@
+# Run `pre-commit install` to set up hooks
+# Run `pre-commit autoupdate` to update hook versions
+repos:
+  - repo: https://github.com/astral-sh/ruff-pre-commit
+    rev: v0.8.0
+    hooks:
+      - id: ruff
+        args: [--fix]
+      - id: ruff-format
+
+  - repo: https://github.com/pre-commit/pre-commit-hooks
+    rev: v5.0.0
+    hooks:
+      - id: trailing-whitespace
+      - id: end-of-file-fixer
+      - id: check-yaml
+      - id: check-added-large-files

{smplfitter-0.2.10 → smplfitter-0.3.2.dev0}/.readthedocs.yaml

@@ -5,8 +5,7 @@ build:
   tools:
     python: "3.10"
   commands:
-    - python -m pip install .
-    - python -m pip install --no-cache-dir -r docs/requirements.txt
+    - python -m pip install ".[docs]"
     - python -m sphinx -E -b html docs $READTHEDOCS_OUTPUT/html
 
 sphinx:

smplfitter-0.3.2.dev0/Makefile

@@ -0,0 +1,22 @@
+.PHONY: install test lint format docs clean
+
+install:
+	pip install -e ".[dev]"
+
+test:
+	pytest -v --cov=smplfitter
+
+lint:
+	ruff check .
+	ruff format --check .
+
+format:
+	ruff check --fix .
+	ruff format .
+
+docs:
+	cd docs && make html
+
+clean:
+	rm -rf build dist *.egg-info .pytest_cache .ruff_cache .coverage
+	find . -type d -name __pycache__ -exec rm -rf {} +

{smplfitter-0.2.10 → smplfitter-0.3.2.dev0}/PKG-INFO

@@ -1,7 +1,7 @@
 Metadata-Version: 2.4
 Name: smplfitter
-Version: 0.2.10
-Summary: Fast inverse kinematics and shape fitting for SMPL-family models in NumPy, PyTorch, and TensorFlow
+Version: 0.3.2.dev0
+Summary: Fast inverse kinematics and shape fitting for SMPL-family models in NumPy, PyTorch, TensorFlow, JAX, and Numba
 Author-email: István Sárándi <istvan.sarandi@uni-tuebingen.de>
 License: MIT License
         
@@ -27,9 +27,26 @@ License: MIT License
         
 Project-URL: Homepage, https://github.com/isarandi/smplfitter
 Project-URL: Repository, https://github.com/isarandi/smplfitter
+Project-URL: Documentation, https://smplfitter.readthedocs.io/
 Project-URL: Issues, https://github.com/isarandi/smplfitter/issues
+Project-URL: Changelog, https://github.com/isarandi/smplfitter/releases
 Project-URL: Author, https://istvansarandi.com
-Requires-Python: >=3.6
+Keywords: smpl,smplx,smplh,3d-human,body-model,ik,inverse-kinematics,parametric-model,pose-estimation,differentiable,gpu,tensorflow,pytorch,jax,numba,numpy,batch-processing
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.8
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Operating System :: OS Independent
+Classifier: Topic :: Scientific/Engineering
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.8
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: numpy
@@ -39,22 +56,29 @@ Provides-Extra: tensorflow
 Requires-Dist: tensorflow; extra == "tensorflow"
 Provides-Extra: pytorch
 Requires-Dist: torch; extra == "pytorch"
-Provides-Extra: keywords
-Requires-Dist: smpl; extra == "keywords"
-Requires-Dist: smplx; extra == "keywords"
-Requires-Dist: smplh; extra == "keywords"
-Requires-Dist: 3d-human; extra == "keywords"
-Requires-Dist: body-model; extra == "keywords"
-Requires-Dist: ik; extra == "keywords"
-Requires-Dist: inverse-kinematics; extra == "keywords"
-Requires-Dist: parametric-model; extra == "keywords"
-Requires-Dist: pose-estimation; extra == "keywords"
-Requires-Dist: differentiable; extra == "keywords"
-Requires-Dist: gpu; extra == "keywords"
-Requires-Dist: tensorflow; extra == "keywords"
-Requires-Dist: pytorch; extra == "keywords"
-Requires-Dist: numpy; extra == "keywords"
-Requires-Dist: batch-processing; extra == "keywords"
+Provides-Extra: jax
+Requires-Dist: jax; extra == "jax"
+Requires-Dist: jaxlib; extra == "jax"
+Provides-Extra: numba
+Requires-Dist: numba; extra == "numba"
+Provides-Extra: test
+Requires-Dist: pytest>=7.0; extra == "test"
+Requires-Dist: pytest-cov; extra == "test"
+Provides-Extra: lint
+Requires-Dist: ruff; extra == "lint"
+Requires-Dist: pre-commit; extra == "lint"
+Provides-Extra: docs
+Requires-Dist: sphinx; extra == "docs"
+Requires-Dist: sphinxcontrib-bibtex; extra == "docs"
+Requires-Dist: sphinx-autoapi; extra == "docs"
+Requires-Dist: sphinx-autobuild; extra == "docs"
+Requires-Dist: sphinx-autodoc-typehints; extra == "docs"
+Requires-Dist: sphinx-codeautolink; extra == "docs"
+Requires-Dist: pydata-sphinx-theme; extra == "docs"
+Requires-Dist: setuptools-scm; extra == "docs"
+Requires-Dist: toml; extra == "docs"
+Provides-Extra: dev
+Requires-Dist: smplfitter[docs,lint,pytorch,test]; extra == "dev"
 Dynamic: license-file
 
 # SMPLFitter: The Fast Way From Vertices to Parametric 3D Humans
@@ -77,11 +101,9 @@ It can fit a batch of 4096 instances in 423 ms on a single RTX 3090 GPU giving a
 ## Installation
 
 ```bash
-pip install git+https://github.com/isarandi/smplfitter.git
+pip install smplfitter
 ```
 
-(Packaging for PyPI is planned for later.)
-
 ### Download Body Model Files
 
 You need to download the body model data files from the corresponding websites for this code to work. You only need the ones that you plan to use. There should be a `DATA_ROOT` environment variable under which a `body_models` directory should look like this:
@@ -111,7 +133,7 @@ $DATA_ROOT/body_models
 └── smplx2smpl_deftrafo_setup.pkl
 ```
 
-You can refer to the relevant [script](https://github.com/isarandi/PosePile/tree/main/posepile/get_body_models.sh) in the PosePile repo about how to download these files.
+You can refer to the relevant [script](https://github.com/isarandi/posepile/tree/main/posepile/src/get_body_models.sh) in the PosePile repo about how to download these files.
 
 ## Usage Examples
 
@@ -121,8 +143,8 @@ You can refer to the relevant [script](https://github.com/isarandi/PosePile/tree
 import torch
 from smplfitter.pt import BodyModel, BodyFitter
 
-body_model = BodyModel('smpl', 'neutral').cuda()  # create the body model to be fitted
-fitter = BodyFitter(body_model, num_betas=10).cuda()  # create the fitter
+body_model = BodyModel('smpl', 'neutral', num_betas=10).cuda()  # create the body model to be fitted
+fitter = BodyFitter(body_model).cuda()  # create the fitter
 fitter = torch.jit.script(fitter)  # optional: compile the fitter for faster execution
 
 # Obtain a batch of nonparametric vertex and joint locations (here we use dummy random data)
@@ -131,7 +153,7 @@ vertices = torch.rand((batch_size, 6890, 3)).cuda()
 joints = torch.rand((batch_size, 24, 3)).cuda()
 
 # Do the fitting!
-fit_res = fitter.fit(vertices, joints, n_iter=3, beta_regularizer=1)
+fit_res = fitter.fit(vertices, joints, num_iter=3, beta_regularizer=1)
 fit_res['pose_rotvecs'], fit_res['shape_betas'], fit_res['trans']
 ```
 
@@ -139,7 +161,7 @@ fit_res['pose_rotvecs'], fit_res['shape_betas'], fit_res['trans']
 
 ```python
 import torch
-from smplfitter.pt import BodyConverter
+from smplfitter.pt import BodyModel, BodyConverter
 
 bm_in = BodyModel('smpl', 'neutral')
 bm_out = BodyModel('smplx', 'neutral')
@@ -158,11 +180,11 @@ out['pose_rotvecs'], out['shape_betas'], out['trans']
 
 ## The Algorithm
 
-SMPL(-X/+H) is a parametric body model that takes body part orientations $\theta$ and body shape vector $\beta$ as inputs and yields vertex and joint locations as outputs. SMPLfit approximates the **inverse operation**: it takes vertex and joint locations as inputs and yields orientations $\theta$ and shape $\beta$ as outputs.
+SMPL(-X/+H) is a parametric body model that takes body part orientations $\theta$ and body shape vector $\beta$ as inputs and yields vertex and joint locations as outputs. SMPLFitter approximates the **inverse operation**: it takes vertex and joint locations as inputs and yields orientations $\theta$ and shape $\beta$ as outputs.
 
 Our algorithm alternates between fitting orientations and fitting shape. A good result can be obtained already with 1-3 iterations.
 
-We illustrate the steps with the following example. Given the depicted RGB image, we used [Neural Localizer Fields (NLF)](https://virtualhumans.mpi-inf.mpg.de/nlf) to obtain nonparametric vertex and joint locations as follows:
+We illustrate the steps with the following example. Given the depicted RGB image, we used [Neural Localizer Fields (NLF)](https://istvansarandi.com/nlf) to obtain nonparametric vertex and joint locations as follows:
 
 <img src="docs/_static/images/image.png" alt="on_image" width="300"/>
 <img src="docs/_static/images/on_image.png" alt="on_image" width="300"/>

{smplfitter-0.2.10 → smplfitter-0.3.2.dev0}/README.md

@@ -18,11 +18,9 @@ It can fit a batch of 4096 instances in 423 ms on a single RTX 3090 GPU giving a
 ## Installation
 
 ```bash
-pip install git+https://github.com/isarandi/smplfitter.git
+pip install smplfitter
 ```
 
-(Packaging for PyPI is planned for later.)
-
 ### Download Body Model Files
 
 You need to download the body model data files from the corresponding websites for this code to work. You only need the ones that you plan to use. There should be a `DATA_ROOT` environment variable under which a `body_models` directory should look like this:
@@ -52,7 +50,7 @@ $DATA_ROOT/body_models
 └── smplx2smpl_deftrafo_setup.pkl
 ```
 
-You can refer to the relevant [script](https://github.com/isarandi/PosePile/tree/main/posepile/get_body_models.sh) in the PosePile repo about how to download these files.
+You can refer to the relevant [script](https://github.com/isarandi/posepile/tree/main/posepile/src/get_body_models.sh) in the PosePile repo about how to download these files.
 
 ## Usage Examples
 
@@ -62,8 +60,8 @@ You can refer to the relevant [script](https://github.com/isarandi/PosePile/tree
 import torch
 from smplfitter.pt import BodyModel, BodyFitter
 
-body_model = BodyModel('smpl', 'neutral').cuda()  # create the body model to be fitted
-fitter = BodyFitter(body_model, num_betas=10).cuda()  # create the fitter
+body_model = BodyModel('smpl', 'neutral', num_betas=10).cuda()  # create the body model to be fitted
+fitter = BodyFitter(body_model).cuda()  # create the fitter
 fitter = torch.jit.script(fitter)  # optional: compile the fitter for faster execution
 
 # Obtain a batch of nonparametric vertex and joint locations (here we use dummy random data)
@@ -72,7 +70,7 @@ vertices = torch.rand((batch_size, 6890, 3)).cuda()
 joints = torch.rand((batch_size, 24, 3)).cuda()
 
 # Do the fitting!
-fit_res = fitter.fit(vertices, joints, n_iter=3, beta_regularizer=1)
+fit_res = fitter.fit(vertices, joints, num_iter=3, beta_regularizer=1)
 fit_res['pose_rotvecs'], fit_res['shape_betas'], fit_res['trans']
 ```
 
@@ -80,7 +78,7 @@ fit_res['pose_rotvecs'], fit_res['shape_betas'], fit_res['trans']
 
 ```python
 import torch
-from smplfitter.pt import BodyConverter
+from smplfitter.pt import BodyModel, BodyConverter
 
 bm_in = BodyModel('smpl', 'neutral')
 bm_out = BodyModel('smplx', 'neutral')
@@ -99,11 +97,11 @@ out['pose_rotvecs'], out['shape_betas'], out['trans']
 
 ## The Algorithm
 
-SMPL(-X/+H) is a parametric body model that takes body part orientations $\theta$ and body shape vector $\beta$ as inputs and yields vertex and joint locations as outputs. SMPLfit approximates the **inverse operation**: it takes vertex and joint locations as inputs and yields orientations $\theta$ and shape $\beta$ as outputs.
+SMPL(-X/+H) is a parametric body model that takes body part orientations $\theta$ and body shape vector $\beta$ as inputs and yields vertex and joint locations as outputs. SMPLFitter approximates the **inverse operation**: it takes vertex and joint locations as inputs and yields orientations $\theta$ and shape $\beta$ as outputs.
 
 Our algorithm alternates between fitting orientations and fitting shape. A good result can be obtained already with 1-3 iterations.
 
-We illustrate the steps with the following example. Given the depicted RGB image, we used [Neural Localizer Fields (NLF)](https://virtualhumans.mpi-inf.mpg.de/nlf) to obtain nonparametric vertex and joint locations as follows:
+We illustrate the steps with the following example. Given the depicted RGB image, we used [Neural Localizer Fields (NLF)](https://istvansarandi.com/nlf) to obtain nonparametric vertex and joint locations as follows:
 
 <img src="docs/_static/images/image.png" alt="on_image" width="300"/>
 <img src="docs/_static/images/on_image.png" alt="on_image" width="300"/>

smplfitter-0.3.2.dev0/benchmark/README.md

@@ -0,0 +1,29 @@
+# Benchmark Results
+
+![Benchmark](benchmark.png)
+
+## Summary (RTX 4090)
+
+**With Vertices (batch=64):**
+| Backend | Throughput |
+|---------|------------|
+| NumPy | 83/sec |
+| Numba | 2,600/sec |
+| PT compile GPU | **64,000/sec** |
+| TF function GPU | 40,000/sec |
+
+**Joints Only (batch=64):**
+| Backend | Throughput |
+|---------|------------|
+| Numba | **152,000/sec** |
+| NumPy | 95,000/sec |
+| PT compile GPU | 69,000/sec |
+
+For joints-only, CPU backends win due to GPU kernel overhead.
+
+## Usage
+
+```bash
+python benchmark/run_benchmark.py  # Generate results.json
+python benchmark/plot_results.py   # Generate plots
+```

smplfitter-0.3.2.dev0/benchmark/bench_bodymodel.py

@@ -0,0 +1,106 @@
+"""Benchmark full body model: NumPy vs Numba vs Cython."""
+
+import numpy as np
+import time
+
+
+def benchmark(func, inputs, n_warmup=3, n_iter=50):
+    """Run benchmark and return mean time in ms."""
+    # Warmup
+    for _ in range(n_warmup):
+        func(**inputs)
+
+    # Timed runs
+    start = time.perf_counter()
+    for _ in range(n_iter):
+        func(**inputs)
+    elapsed = time.perf_counter() - start
+    return elapsed / n_iter * 1000  # ms
+
+
+def main():
+    import smplfitter.np as smpl_np
+    import smplfitter.nb as smpl_nb
+
+    try:
+        import smplfitter.cy.bodymodel as smpl_cy
+
+        has_cython = True
+    except ImportError as e:
+        print(f'Cython not built: {e}')
+        print('Run: python setup_cy.py build_ext --inplace')
+        has_cython = False
+
+    print('Benchmarking BodyModel forward pass')
+    print('=' * 70)
+
+    for model_name in ['smpl']:
+        print(f'\nModel: {model_name.upper()}')
+
+        model_np = smpl_np.BodyModel(model_name=model_name, num_betas=10)
+        model_nb = smpl_nb.BodyModel(model_name=model_name, num_betas=10)
+        if has_cython:
+            model_cy = smpl_cy.BodyModel(model_name=model_name, num_betas=10)
+
+        num_joints = model_np.num_joints
+
+        for return_vertices in [False, True]:
+            mode = 'with_vertices' if return_vertices else 'joints_only'
+            print(f'\n  Mode: {mode}')
+
+            for batch_size in [1, 8, 32, 128, 512]:
+                pose = np.random.randn(batch_size, num_joints * 3).astype(np.float32) * 0.1
+                shape = np.random.randn(batch_size, 10).astype(np.float32) * 0.5
+                trans = np.random.randn(batch_size, 3).astype(np.float32)
+
+                inputs = dict(
+                    pose_rotvecs=pose,
+                    shape_betas=shape,
+                    trans=trans,
+                    return_vertices=return_vertices,
+                )
+
+                time_np = benchmark(model_np, inputs)
+                time_nb = benchmark(model_nb, inputs)
+
+                print(f'\n    batch={batch_size:3d}')
+                print(
+                    f'      NumPy:  {time_np:8.3f} ms  ({batch_size/time_np*1000:8.0f} items/sec)'
+                )
+                print(
+                    f'      Numba:  {time_nb:8.3f} ms  ({batch_size/time_nb*1000:8.0f} items/sec)  {time_np/time_nb:5.1f}x vs NumPy'
+                )
+
+                if has_cython:
+                    time_cy = benchmark(model_cy, inputs)
+                    print(
+                        f'      Cython: {time_cy:8.3f} ms  ({batch_size/time_cy*1000:8.0f} items/sec)  {time_np/time_cy:5.1f}x vs NumPy, {time_nb/time_cy:5.2f}x vs Numba'
+                    )
+
+                # Verify correctness
+                result_np = model_np(**inputs)
+                result_nb = model_nb(**inputs)
+
+                if not np.allclose(result_np['joints'], result_nb['joints'], atol=1e-4):
+                    print('      WARNING: Numba joints differ!')
+                if return_vertices and not np.allclose(
+                    result_np['vertices'], result_nb['vertices'], atol=1e-4
+                ):
+                    print('      WARNING: Numba vertices differ!')
+
+                if has_cython:
+                    result_cy = model_cy(**inputs)
+                    if not np.allclose(result_np['joints'], result_cy['joints'], atol=1e-4):
+                        print('      WARNING: Cython joints differ!')
+                        diff = np.abs(result_np['joints'] - result_cy['joints']).max()
+                        print(f'               Max diff: {diff}')
+                    if return_vertices and not np.allclose(
+                        result_np['vertices'], result_cy['vertices'], atol=1e-4
+                    ):
+                        print('      WARNING: Cython vertices differ!')
+                        diff = np.abs(result_np['vertices'] - result_cy['vertices']).max()
+                        print(f'               Max diff: {diff}')
+
+
+if __name__ == '__main__':
+    main()