🎉 Celebrating 25 Years of GameDev.net! 🎉

Not many can claim 25 years on the Internet! Join us in celebrating this milestone. Learn more about our history, and thank you for being a part of our community!

Back to Graphics and GPU Programming

Transporting a float16 Image (OpenEXR Float (Half)) to HLSL Shader

Flub · 2021-09-02T06:36:09

Hi there, I am a newbie, so please excuse if I dont know the right words for things… For my Project I rendered an image in OpenEXR Half (Float) (16bit Float) Format (described here). The black and white image is part of an animation sequence and is one of two parts of a composition step which i want to do in real time through an HLSL Shader. Now, as I want to play a Video through the HLSL Shader I somehow need to pack/encode the 16bit Float Value into the R and the G Channels of and RGB (8bit per channel) image and then decode it in the HLSL Shader. When i just export the 16bit float image to 8bit png I get the following disaapointing result: When i multiply the 16bit float image with the color, it looks good. Thats how i want it to come out of the HLSL Shader: To generate the 8bit PNGS i use a special Blender Version which offers the possibility to use a python script in a compositor node.This way i get the image in python as float32 array. import bpy # If you want to read any Blender data import numpy as np import pycompositor # Helper script from PIL import Image, ImageDraw # Pillow import sys def on_main(context): """on_main() is called in the main thread and can be used to access Blender and other libraries in a thread-safe manner. It is possible to add extra data to the context object and use it later in the on_async().""" images = context["images"] # A list of 4 node input images inputs = context["inputs"] # A list of 8 node input values outputs = context["outputs"] # A list of 1 node output image. Might have more in the future out = outputs[0] # Final output image # Copy Image 0 to output without any modifications #out[:] = images[0] # Set output to green #out[:] = (0, 1, 0, 1) # ################### HERE <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<< out[:] = images[0] # This is the image as float32 type numpy array out[:,:,0] = (images[0][:,:,0].astype("int16") >> 8) & 0xff # This is what i tried to split the value out[:,:,1] = images[0][:,:,0].astype("int16") & 0xff def on_async(context): """on_async() is called in background thread after on_main() and should be used for processing that takes a long time in order to avoid blocking Blender UI. Do not touch anything that is not thread safe from this function.""" pass Can someone please help me with the two steps. First how to encode the original float16 image data in the R and G channels of a png and output that. Second, what to write into the shader to get that float again out of R and G for the multiplication with the color. This is my shader so far: uniform float4 color; uniform int2 builtin_uv_size; // the source dimensions uniform string notes = "This custom Shader is used to colorize Florian Deeg's coffee machine animation. It expects double Frame image in the format 3840x1080 where the left one is a b/w image that needs to be multiplied with a color and the right one the environment which needs to be added to that."; // This function returns pixel color at x,y float4 get(VertData v_in, float x, float y) { return image.Sample( textureSampler, float2( v_in.uv.x + x * uv_pixel_interval.x, v_in.uv.y + y * uv_pixel_interval.y ) ); } float4 mainImage(VertData v_in) : TARGET { float4 rgba = image.Sample(textureSampler, v_in.uv); float colorIntensity; float colorIntensityUpper = rgba.b; float colorIntensityLower = rgba.g; colorIntensity = colorIntensityUpper *2*2*2*2*2*2*2*2 + colorIntensityLower; rgba = color * colorIntensity; //rgba.g = color * 0; //rgba.b = color * 0; // this is the upper byte of the half float value return rgba; } Thank you so much for any help, I am really frustrated because I could not get it done for days now and I need it for my exam. -Florian