Amélie Heinrich @Dispatch_Graph

WHAT THE FUCK

meow @reallybigtail

a day ago

THEY FOUND A GREEN STAR?????/

28 32 1K 876K 91

society does not value art enough. we should be using modern tech to empower artists to build shit that wae never possible before. i want to leave my house and think "woah im the future".

lost in latency @lostnlatency

3 days ago

the 2000s thought airports would have fish tanks

115 2K 22K 953K 5K

MicroSDF: Microfacet-Driven Hybrid Neural SDFs for Mixed-Reflectance Surface Reconstruction Lejia Ye, Yuhua Xu, Yulan Guo, Lian Xu, Xiaohu Zhang arxiv.org/abs/2610.19681 Abstract: Accurate 3D reconstruction in real-world environments remains a significant challenge due to the coexistence of reflective and non-reflective surfaces, which pose distinct modeling demands. Existing methods often treat these surface types separately, limiting their generalizability and physical plausibility. To bridge this gap, we propose MicroSDF, a novel neural implicit framework that facilitates geometry and reflectance modeling through microfacet theory. Our approach incorporates three core innovations: (i) a microfacet-guided geometry model that extracts multi-scale surface normals (macroscopic and microfacet) from a signed distance field (SDF), regularized by a proposed microfacet normal consistency loss to enforce physically plausible surface orientations; (ii) an enhanced dual-branch color model, where the specular branch leverages the microfacet normals to model high-frequency reflectance, and the vanilla branch, unlike prior works, uses reflection direction (instead of viewing direction) to better model diffuse and low-frequency specular components; and (iii) a detection-guided color blending strategy that adaptively fuses the color outputs based on reflection priors, providing more physically intuitive blending than implicitly learned blending weights. Combined with a tailored multi-stage optimization scheme, the proposed MicroSDF achieves robust and high-fidelity reconstruction across reflective and non-reflective surfaces. Extensive experiments on DTU, Shiny Blender, Ref-NeRF, and DeepVoxels datasets demonstrate state-of-the-art performance, establishing a new direction for physically grounded neural reconstruction.

keyboard_arrow_left Previous keyboard_arrow_right Next

As a computer graphics person, M3 goes in S tier. It was the first Apple Silicon chip to gain dynamic caching and hardware ray tracing, and Apple’s implementations of both of those features is really good.

Cinq @WhosCinq

a week ago

ranking all M-chips ever released

244 216 10K 841K 1K

@frguthmann Bindless as in proper HLSL ResourceDescriptorHeap? How do they handle barriers then?

@SolidBasis No native Slang debugging for Mac. Also not a big fan of the macro hell you have to do to get cross platform bindless using Slang, with HLSL+DXC it's a loooot easier

If you are a graphics programmer making your own engine, I'd like to ask a question. Currently you might be using graphics API X or Y or Z or whatever. Take a look at AGFX (link in comments) -- what's one thing that would stop you from using it?

@SolidBasis - For heap allocations, it was just easier for me to keep things opaque. I understand the criticism though - Same for RT pipeline support. (It's honestly quite the hassle with Metal shader converter so I decided not to do it)

@leobenaducci Forgot to add it, but its MIT

@gh8472 AGFX vs SteamDB Hardware Report, percentage of users that you can target: macOS (M1+): 80.6% Windows (SM6.6): 85-90% (estimate because enhanced barriers isnt a hardware feature that steamdb reports) Linux: 80%, steam deck compatible

@gaynwatch Sadly it was the price to pay to get Metal 4. But my logic is -- AGFX is rather new, by the time people ship games with it, most people will have switched to 26 by now

@hugoamnov D3D12 support so potential Xbox support later down the line. Since I use mutable descriptor instead of descriptor heap I can target more hardware as well at the cost of a more older looking API

@RyDawgE_ Fair

@FlyingStunt Not sure why having multi draw indirect support marker in the device info be a good reason to just not use agfx. And yes Vulkan requires mutable descriptor but no real way around that. Mutable descriptor has better support than descriptor heap

@milostosic Can BGFX do hardware raytracing? Multi draw indirect? Lightweight shader compiler? Bindless? BGFX is great for legacy hardware. AGFX is meant for higher end

@engineseer I used xmake purely for its package manager so its easier for the demo, but adding your own CMakeLists.txt for the project should be dead simple, only requires a few platform links (Foundation, Metal, dxgi etc), should be in the README

@cyannick What AGFX has: - GPU timings via timestamp queries - No VRS as Metal doesn't support it - D3D12 support - No VR support yet as I don't have a headset, but shouldn't be too hard to add

@stephc_int13 You might need agility SDK. I didn't ship it in the AIMD demo so it's definitely an oversight. Does the AGFX demos work for you otherwise?

@mikulasflorek AGFX is almost 1 on 1 with target API with 0 overhead target, raytracing, mesh shaders, multi draw indirect, bindless support, and supports D3D12/Vulkan/Metal 4. Main target is indies or AA

@mikulasflorek I haven't found much when I did my research: - BGFX targets lower level APIs at the cost of less features - Haven't used The Forge yet - SDL3_GPU is basically the same as BGFX - NVRHI has no Metal support

@stephc_int13 You need the graphics development toolkit from Windows (should be able to install it from developer settings) to get D3D12 debug validation. Otherwise, you can just turn off enableValidation when creating the AGFX device, should run fine