Merge pull request #45 from rnd-team-dev/r0.16.0

R0.16.0

Author: robertsulej

CHANGELOG.rst

@@ -1,6 +1,20 @@
 Release history
 ===============
 
+`v0.16.0` - 2023-05-17
+----------------------
+
+Added
+~~~~~
+
+- set texture data from pytorch tensors
+- texture filter modes enabled, default is *trilinear*, optionally can be changed to *nearest*
+
+- set geometry data directly from numpy array or pytorch tensor (no host copy is mada)
+- synchronization method to update host copy with data from gpu: NpOptiX.sync_raw_data()
+
+- convenience method to get a copy of geometry data (more simple than PinnedBuffer, though returns a copy): NpOptiX.get_data()
+
 `v0.15.1` - 2023-03-16
 ----------------------
 
@@ -568,6 +582,7 @@ Added
 - this changelog, markdown description content type tag for PyPI
 - use [Semantic Versioning](https://semver.org/spec/v2.0.0.html)
 
+.. _`v0.16.0`: https://github.com/rnd-team-dev/plotoptix/releases/tag/v0.16.0
 .. _`v0.15.1`: https://github.com/rnd-team-dev/plotoptix/releases/tag/v0.15.1
 .. _`v0.15.0`: https://github.com/rnd-team-dev/plotoptix/releases/tag/v0.15.0
 .. _`v0.14.4`: https://github.com/rnd-team-dev/plotoptix/releases/tag/v0.14.4

README.rst

@@ -15,13 +15,14 @@ PlotOptiX
 
 `Docs <https://plotoptix.rnd.team>`__
 
-- Check what we are doing with PlotOptiX on `Behance <https://www.behance.net/RnDTeam>`__, `Facebook <https://www.facebook.com/rndteam>`__, and `Instagram <https://www.instagram.com/rnd.team.studio/>`__.
+- Have a look what is possible with PlotOptiX: `Behance <https://www.behance.net/RnDTeam>`__, `Instagram <https://www.instagram.com/rnd.team.studio/>`__, and `Facebook <https://www.facebook.com/rndteam>`__.
 - Join us on `Patreon <https://www.patreon.com/rndteam?fan_landing=true>`__ for news, release plans and hi-res content.
 
 PlotOptiX is a 3D `ray tracing <https://en.wikipedia.org/wiki/Ray_tracing_(graphics)>`__ package for Python, aimed at easy and aesthetic visualization
 of large datasets (and small as well). Data features can be represented in images as a position, size/thickness and color of primitives
 of several basic shapes, or projected onto surfaces of objects in form of a color textures and displacement maps. Triangular meshes,
-generated in the code or loaded from a file, are supported as well. All is finished with a photorealistic lighting, depth of field, and many other physically based effects simulated with a high quality.
+generated in the code or loaded from a file, are supported as well. All is finished with a photorealistic lighting, depth of field, and many other
+physically based effects simulated with a high quality.
 
 No need to write shaders, intersection algorithms, handle 3D scene technicalities. Basic usage is even more simple than with
 `matplotlib <https://matplotlib.org/gallery/mplot3d/scatter3d.html>`__:
@@ -65,8 +66,9 @@ Features
 - *light sources*: spherical and parallelogram, light emission in volumes, uniform environmental light or environment map
 - *post-processing*: tonal correction curves, levels adjustment, apply mask/overlay, AI denoiser and upsampler
 - *callbacks*: at the scene initialization, start and end of each frame raytracing, end of progressive accumulation
-- 8/16/32bps(hdr) image output to `numpy <http://www.numpy.org>`__ array, or save to popular image file formats
+- 8/16/32bps(hdr) image output to `numpy <https://numpy.org>`__ array, or save to popular image file formats
 - zero-copy access to GPU buffers wrapped in ndarrays: 8/32bpc image, hit and object info, albedo, normals
+- direct access to `PyTorch <https://pytorch.org>`__ tensors data stored on GPU (and CPU as well) for texture and geometry updates
 - GPU acceleration using RT Cores and everything else what comes with `OptiX <https://developer.nvidia.com/optix>`__
 - hardware accelerated video output to MP4 file format using `NVENC 9.0 <https://developer.nvidia.com/nvidia-video-codec-sdk>`__
 - Tkinter based simple GUI window or a headless raytracer
@@ -95,7 +97,7 @@ What's Included
 Installation
 ============
 
-**Note**, at this point, PlotOptiX binaries are tested in: Windows 10/11 (any Python 3), and Linux (Python 3.8 recommended): Ubuntu 18.04, CentOS 7.
+**Note**, at this point, PlotOptiX binaries are tested in: Windows 10/11 (any Python 3), and Linux (Python 3.8-3.10 recommended): Ubuntu 18.04, CentOS 7.
 
 PlotOptiX was also successfully tested on the `Google Cloud Platform <https://cloud.google.com/>`__, using Compute Engine instance with 2x V100 GPU's and Ubuntu 18.04 image.
 Here are the `installation steps <https://github.com/rnd-team-dev/plotoptix/blob/master/gcp_install_gpu.txt>`__ so you can save some precious seconds (FFmpeg not included).
@@ -107,15 +109,7 @@ Windows prerequisites
 
 *.NET Framework:*
 
-Most likely you already got the right version with your Windows installation. Just in case of doubts, here is the command verifying this::
-
-   C:\>reg query "HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\NET Framework Setup\NDP\v4\full" /v version
-   
-   HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\NET Framework Setup\NDP\v4\full
-       version    REG_SZ    4.8.04084
-
-If the number in your output is < 4.8, visit `download page <https://dotnet.microsoft.com/download/dotnet-framework>`__ and
-install the most recent release.
+You have it built in your Windows. Go ahead and install PlotOptiX.
 
 Linux prerequisites
 -------------------
@@ -191,21 +185,14 @@ Then, try running code from the top of this readme, or one of the examples. You
 Development path
 ================
 
-This is still an early version. There are some important features not available yet, eg. ticks and labels on plot axes.
+This is still an experimental version in many aspects.
 
 PlotOptiX is basically an interface to RnD.SharpOptiX library which we are developing and using in our Studio. RnD.SharpOptiX offers
-much more functionality than it is now available through PlotOptiX. We'll progressively add more to PlotOptiX if there is interest in
-this project (download, star, and `become our Patron <https://www.patreon.com/rndteam>`__
-if you like it!).
-
-The idea for development is:
-
-1. Binaries for Linux (done in v0.3.0).
-2. Migrate to OptiX 7.0 (done in v0.7.0).
-3. Complete the plot layout / cover more raytracing features.
-4. Convenience functions for various plot styles. Other GUI's.
+much more functionality than it is now available through PlotOptiX. We progressively add more to PlotOptiX based mostly on the interest
+of our patrons and applications that this project supports.
 
-   *Here, the community input is possible and warmly welcome!*
+Download, star, and `become our Patron <https://www.patreon.com/rndteam>`__ if you like the project. Get in touch, share your use case,
+we are always happy to help and take part in exciting ideas of our users.
 
 Examples
 ========

docs/npoptix.rst

@@ -25,7 +25,7 @@ basic rules are:
 .. image:: images/flow_1.png
    :alt: PlotOptiX compute flow
 
-Fig. 1. PlotOptiX computations flow. Details of the OptiX tast are ommited for clarity (i.e. scene compilation and multi-gpu management).
+Fig. 1. PlotOptiX computations flow. Details of the OptiX task are ommited for clarity (i.e. scene compilation and multi-gpu management).
 
 .. image:: images/flow_2.png
    :alt: PlotOptiX compute flow

docs/npoptix_geometry.rst

@@ -1,10 +1,11 @@
-Plot geometry
-=============
+Geometry
+========
 
 Create, load, update, and remove a plot
 ---------------------------------------
 
 .. automethod:: plotoptix.NpOptiX.get_geometry_names
+.. automethod:: plotoptix.NpOptiX.get_data
 .. automethod:: plotoptix.NpOptiX.set_data
 .. automethod:: plotoptix.NpOptiX.update_data
 .. automethod:: plotoptix.NpOptiX.set_data_2d
@@ -19,10 +20,17 @@ Create, load, update, and remove a plot
 .. automethod:: plotoptix.NpOptiX.load_displacement
 .. automethod:: plotoptix.NpOptiX.delete_geometry
 
-Direct modifications of data
-----------------------------
+Direct access and modifications of data
+---------------------------------------
+
+**Fast updates:**
+
+.. automethod:: plotoptix.NpOptiX.update_raw_data
+.. automethod:: plotoptix.NpOptiX.sync_raw_data
+
+**Geometry modifications:**
 
-These methods allow making changes to properties of data points
+Following methods allow making changes to properties of data points
 stored internally in the raytracer, without re-sending whole data
 arrays from the Python code.
 

examples/2_animations_and_callbacks/5_pytorch_texture.py

@@ -0,0 +1,73 @@
+"""
+Pytorch texture source - game of life.
+"""
+
+import torch
+import torch.nn.functional as F
+
+from plotoptix import TkOptiX
+from plotoptix.materials import m_flat
+
+
+class params():
+    if torch.cuda.is_available:
+        device = torch.device('cuda')
+        dtype = torch.float16
+    else:
+        device = torch.device('cpu')
+        dtype = torch.float32
+    w = torch.tensor(
+        [[[[1.0,1.0,1.0], [1.0,0.0,1.0], [1.0,1.0,1.0]]]],
+        dtype=dtype, device=device, requires_grad=False
+    )
+    cells = torch.rand((1,1,128,128), dtype=dtype, device=device, requires_grad=False)
+    cells[cells > 0.995] = 1.0
+    cells[cells < 1.0] = 0.0
+    tex2D = torch.unsqueeze(cells[0, 0].type(torch.float32), -1).expand(-1, -1, 4).contiguous()
+
+
+# Update texture data with a simple "game of life" rules.
+def compute(rt, delta):
+    params.cells = F.conv2d(params.cells, weight=params.w, stride=1, padding=1)
+    params.cells[params.cells < 2.0] = 0.0
+    params.cells[params.cells > 3.0] = 0.0
+    params.cells[params.cells != 0.0] = 1.0
+
+    # Conversion to float32 and to contiguous memoty layout is ensured by plotoptix,
+    # though you may wamt to make it explicit like here, eg for performance reasons.
+    params.tex2D = torch.unsqueeze(params.cells[0, 0].type(torch.float32), -1).expand(-1, -1, 4).contiguous()
+
+
+# Copy texture data to plotoptix scene.
+def update_data(rt):
+    rt.set_torch_texture_2d("tex2d", params.tex2D, refresh=True)
+
+
+def main():
+    rt = TkOptiX(
+        on_scene_compute=compute,
+        on_rt_completed=update_data
+    )
+    rt.set_param(min_accumulation_step=1)
+    rt.set_background(0)
+    rt.set_ambient(0)
+
+    # NOTE: pytorch features are not enabled by default. You need
+    # to call this method before using anything related to pytorch.
+    rt.enable_torch()
+
+    rt.set_torch_texture_2d("tex2d", params.tex2D, addr_mode="Clamp", filter_mode="Nearest")
+    m_flat["ColorTextures"] = [ "tex2d" ]
+    rt.setup_material("flat", m_flat)
+
+    rt.setup_camera("cam1", eye=[0, 0, 3], target=[0, 0, 0], fov=35, glock=True)
+
+    rt.set_data("plane", geom="Parallelograms", mat="flat",
+                pos=[-1, -1, 0], u=[2, 0, 0], v=[0, 2, 0], c=0.9
+    )
+
+    rt.start()
+
+if __name__ == '__main__':
+    main()
+

examples/2_animations_and_callbacks/6_direct_data_update.py

@@ -0,0 +1,64 @@
+"""
+Direct update of geometry data on device.
+
+This example uses numpy arrays, but pytorch tensors are supported as well,
+just note that rt.enable_torch() call is required before using pytorch.
+"""
+
+import numpy as np
+
+from plotoptix import TkOptiX
+from plotoptix.utils import simplex
+
+
+class params():
+    n = 100
+    rx = (-20, 20)
+    r = 0.85 * 0.5 * (rx[1] - rx[0]) / (n - 1)
+
+    x = np.linspace(rx[0], rx[1], n)
+    z = np.linspace(rx[0], rx[1], n)
+    X, Z = np.meshgrid(x, z)
+
+    data = np.stack([X.flatten(), np.zeros(n*n), Z.flatten()], axis=1)
+    t = 0
+    
+# Compute updated geometry data.
+def compute(rt, delta):
+    row = np.ones((params.data.shape[0], 1))
+    xn = simplex(np.concatenate([params.data, params.t * row], axis=1))
+    yn = simplex(np.concatenate([params.data, (params.t + 20) * row], axis=1))
+    zn = simplex(np.concatenate([params.data, (params.t - 20) * row], axis=1))
+    dv = np.stack([xn, yn, zn], axis=1)
+    params.data += 0.02 * dv
+    params.t += 0.05
+    
+# Fast copy to geometry buffer on device, without making a host copy.
+def update_data(rt):
+    rt.update_raw_data("balls", pos=params.data)
+
+
+def main():
+    rt = TkOptiX(
+        on_scene_compute=compute,
+        on_rt_completed=update_data
+    )
+    rt.set_param(min_accumulation_step=10, max_accumulation_frames=16)
+    rt.set_background(0)
+    rt.set_ambient(0)
+
+    rt.setup_camera("cam1", cam_type="ThinLens", eye=[3.5, 1.27, 3.5], target=[0, 0, 0], fov=30, glock=True)
+    rt.setup_light("light1", pos=[4, 5, 5], color=18, radius=1.0)
+
+    exposure = 1.0; gamma = 2.2
+    rt.set_float("tonemap_exposure", exposure)
+    rt.set_float("tonemap_gamma", gamma)
+    rt.add_postproc("Gamma")
+
+    rt.set_data("balls", pos=params.data, c=0.82, r=params.r)
+
+    rt.start()
+
+if __name__ == '__main__':
+    main()
+

plotoptix/__init__.py

@@ -12,8 +12,8 @@
 
 __author__  = "Robert Sulej, R&D Team <dev@rnd.team>"
 __status__  = "beta"
-__version__ = "0.15.1"
-__date__    = "16 March 2023"
+__version__ = "0.16.0"
+__date__    = "17 May 2023"
 
 import logging