RT @twominutepapers: NVIDIA’s AI watched 150,000 videos… and learned to relight scenes incredibly well! No game engine. No 3D software. And….

Following DiffusionRenderer, we made another attempt towards general-purpose relighting ✨. It can address some challenging rendering effects beyond DiffusionRenderer's capability. Please check @Kai__He's thread and webpage for more examples 🚀🚀🚀.

Join millions who have switched to Grok.

RT @sopharicks: Excited to host @ZGojcic talk next week on @nvidia DiffusionRenderer, a new technique for neural rendering. It approximate….

RT @zianwang97: 🚀 We just open-sourced Cosmos DiffusionRenderer!. This major upgrade brings significantly improved video de-lighting and re….

Thanks for the invitation from @anagh_malik, DiffusionRenderer will also be presented in the poster session of the Pi3DVI Workshop ( tomorrow afternoon.

Excited to announce our code release for DiffusionRenderer! Hope you have fun with it 😎. Catch our oral and poster presentations this Saturday afternoon at #CVPR25, Please feel free to visit!.

RT @JunGao33210520: This year, we have 3 papers in CVPR, discussing the connection between 3D and video models:. GEN3C [Highlight] 3D groun….

RT @Dazitu_616: 📢 Introducing DenseDPO: Fine-Grained Temporal Preference Optimization for Video Diffusion Models. Compared to vanilla DPO,….

Please check @ZhiHaoLin16 's weather magic🤓.

RT @xuanchi13: Thanks @_akhaliq for sharing our GEN3C! GEN3C can easily be applied to creating video/scene from a single image, sparse-view….

DiffusionRenderer reformulates the traditional rendering process as a generative process. I am proud to be one of the main contributors to this exciting project!.Hope one day this technique can be applied to game and movie production🤓.

This project was done with an amazing team at @NVIDIAAI @UofT by @RfLiang @ZGojcic @merlin_ND @davidjesusacu @nanditav17 @FidlerSanja @zianwang97 . Find more details on our project page:

The application of DiPIR can extend beyond lighting estimation – we show that the diffusion model can also propagate text-condition to material editing (e.g., base color, roughness, and emission), or adjust tone-mapping curves.

Benefiting from the physically-based scene representation, we can consistently insert virtual objects across multiple viewpoints.

The optimized results can be used to create photorealistic editing, such as animating inserted objects with either a moving background or a moving object.

DiPIR uses a differentiable path tracer to simulate light interactions with inserted objects. The diffusion model is personalized based on input image and inserted asset. An adapted SDS provides gradient guidance, which is then backpropagated through the differentiable renderer.

🔍 In DiPIR, the diffusion model acts like a human evaluator. It takes the edited image as input and propagates the feedback signal to physically-based scene attributes via differentiable rendering, enabling end-to-end optimization for attributes such as light and tone-mapping.

🌟 Seamlessly insert 3D objects into any scene!. 🚀 Introducing our #ECCV2024 work “DiPIR”, a method to recover lighting from a single image, enabling photorealistic virtual object compositing into indoor and outdoor scenes. Project page:

Our paper and codes are available here: I will also have an oral presentation and a poster session on Wednesday morning. I'm looking forward to exciting discussions in Paris!.(4/n)

We also do the second round of ray-marching along surface-reflected camera rays to synthesize interreflections for glossy surfaces. (3/n)