Lotus: Diffusion-based Visual Foundation Model for High-quality Dense Prediction
Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel pith:SQ4YSL2Arecord.jsonopen to challenge →
read the original abstract
Leveraging the visual priors of pre-trained text-to-image diffusion models offers a promising solution to enhance zero-shot generalization in dense prediction tasks. However, existing methods often uncritically use the original diffusion formulation, which may not be optimal due to the fundamental differences between dense prediction and image generation. In this paper, we provide a systemic analysis of the diffusion formulation for the dense prediction, focusing on both quality and efficiency. And we find that the original parameterization type for image generation, which learns to predict noise, is harmful for dense prediction; the multi-step noising/denoising diffusion process is also unnecessary and challenging to optimize. Based on these insights, we introduce Lotus, a diffusion-based visual foundation model with a simple yet effective adaptation protocol for dense prediction. Specifically, Lotus is trained to directly predict annotations instead of noise, thereby avoiding harmful variance. We also reformulate the diffusion process into a single-step procedure, simplifying optimization and significantly boosting inference speed. Additionally, we introduce a novel tuning strategy called detail preserver, which achieves more accurate and fine-grained predictions. Without scaling up the training data or model capacity, Lotus achieves SoTA performance in zero-shot depth and normal estimation across various datasets. It also enhances efficiency, being significantly faster than most existing diffusion-based methods. Lotus' superior quality and efficiency also enable a wide range of practical applications, such as joint estimation, single/multi-view 3D reconstruction, etc. Project page: https://lotus3d.github.io/.
This paper has not been read by Pith yet.
Forward citations
Cited by 23 Pith papers
-
TrackCraft3R: Repurposing Video Diffusion Transformers for Dense 3D Tracking
TrackCraft3R is the first method to repurpose a video diffusion transformer as a feed-forward dense 3D tracker via dual-latent representations and temporal RoPE alignment, achieving SOTA performance with lower compute.
-
MUSE: Unlocking Timestep as Native Task Steering for One-Step Dense Prediction
MUSE shows that the native timestep embedding in diffusion models acts as a parameter-free steering signal for multi-task monocular depth and normal estimation via manifold decoupling in latent space.
-
Image Generators are Generalist Vision Learners
An image generator is instruction-tuned to perform diverse vision tasks by representing task outputs as RGB images, achieving SOTA on segmentation and depth estimation.
-
CDPR: Cross-modal Diffusion with Polarization for Reliable Monocular Depth Estimation
CDPR integrates polarization priors into a diffusion-based monocular depth estimator via shared latent space and adaptive gating, outperforming RGB-only methods in challenging scenes.
-
How to Spin an Object: First, Get the Shape Right
Camera-Relative Object Coordinates (CROCS) as an intermediate geometry representation in two-stage image-to-3D models yields superior novel-view quality, geometric accuracy, and multiview consistency over depth maps, ...
-
PointDiT: Pixel-Space Diffusion for Monocular Geometry Estimation
PointDiT is a from-scratch pixel-space Diffusion Transformer for monocular 3D point map estimation that outperforms latent diffusion models in sharpness and ambiguous regions while using a simpler architecture.
-
UniGP: Taming Diffusion Transformer for Prior-Preserved Unified Generation and Perception
UniGP unifies controllable generation and dense prediction in an MMDiT-based diffusion model through simple joint training that preserves backbone priors.
-
AerialMetric: Benchmarking and Adapting UAV Monocular Metric Depth Estimation in the Real World
AerialMetric is a new benchmark dataset and evaluation suite for adapting monocular metric depth estimation models to real-world UAV aerial views.
-
Error-Conditioned Neural Solvers
Error-Conditioned Neural Solvers improve PDE prediction accuracy by using the residual field as network input for learned corrections, outperforming residual-minimization methods by up to 10x on turbulent flows and ge...
-
Modality Forcing for Scalable Spatial Generation
Modality Forcing lets a single DiT produce image and depth outputs in any order after training on sparse real-world depth, with larger image-pretrained models yielding better depth accuracy and a 57% AbsRel reduction ...
-
Open-Source Image Editing Models Are Zero-Shot Vision Learners
Open-source image-editing models show competitive zero-shot performance on monocular depth, surface normals, and semantic segmentation, sometimes matching tuned models.
-
UniVidX: A Unified Multimodal Framework for Versatile Video Generation via Diffusion Priors
UniVidX unifies diverse video generation tasks into one conditional diffusion model using stochastic condition masking, decoupled gated LoRAs, and cross-modal self-attention.
-
Diffusion Model as a Generalist Segmentation Learner
DiGSeg repurposes diffusion U-Nets as generalist segmentation learners by conditioning on image-mask latents and multi-scale CLIP text features, achieving strong cross-domain performance.
-
Image Generators are Generalist Vision Learners
Image generation pretraining produces generalist vision models that reframe perception tasks as image synthesis and reach SOTA results on segmentation, depth estimation, and other 2D/3D tasks.
-
Image Generators are Generalist Vision Learners
Image generation pretraining builds generalist vision models that reach SOTA on 2D and 3D perception tasks by reframing them as RGB image outputs.
-
Need for Speed: Zero-Shot Depth Completion with Single-Step Diffusion
Marigold-SSD delivers zero-shot depth completion via single-step diffusion with late fusion, achieving fast inference after only 4.5 GPU days of training while showing strong cross-domain results on indoor and outdoor...
-
Rolling Sink: Bridging Limited-Horizon Training and Open-Ended Testing in Autoregressive Video Diffusion
Rolling Sink is a training-free cache adjustment technique that maintains visual consistency in autoregressive video diffusion models for ultra-long open-ended generation beyond training horizons.
-
Lotus-2: Advancing Geometric Dense Prediction with Powerful Image Generative Model
Lotus-2 is a two-stage deterministic adaptation of diffusion priors that achieves state-of-the-art monocular depth estimation with only 59K training samples.
-
Ouroboros: Single-step Diffusion Models for Cycle-consistent Forward and Inverse Rendering
Ouroboros uses two single-step diffusion models with cycle consistency for forward and inverse rendering, extending intrinsic decomposition to indoor/outdoor scenes with faster inference than multi-step methods.
-
VolFill: Single-View Amodal 3D Scene Reconstruction with Volumetric Flow Matching
VolFill uses a hybrid 3D VAE to compress sparse truncated unsigned distance function grids into latent space and a latent Diffusion Transformer to denoise complete scenes, conditioned on geometry foundation models, ou...
-
Towards Consistent Video Geometry Estimation
ViGeo is a feed-forward transformer for video geometry that introduces dynamic chunking attention and a completion-based data refinement framework to achieve SOTA on depth, normals, and point map estimation.
-
The Midas Touch for Metric Depth
MTD turns relative depth into metric depth via segment-wise sparse graph optimization and discontinuity-aware geodesic pixel refinement, claiming better accuracy and generalization than prior depth methods.
-
DepthMaster: Taming Diffusion Models for Monocular Depth Estimation
DepthMaster proposes a single-step diffusion model with Feature Alignment and Fourier Enhancement modules in a two-stage training process to improve generalization and detail preservation in monocular depth estimation...
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.