GRIN: Zero-Shot Metric Depth with Pixel-Level Diffusion

TL;DR

This paper presents GRIN, an efficient diffusion model designed to ingest sparse unstructured training data, and uses image features with 3D geometric positional encodings to condition the diffusion process both globally and locally, generating depth predictions at a pixel-level.

Abstract

3D reconstruction from a single image is a long-standing problem in computer vision. Learning-based methods address its inherent scale ambiguity by leveraging increasingly large labeled and unlabeled datasets, to produce geometric priors capable of generating accurate predictions across domains. As a result, state of the art approaches show impressive performance in zero-shot relative and metric depth estimation. Recently, diffusion models have exhibited remarkable scalability and generalizable properties in their learned representations. However, because these models repurpose tools originally designed for image generation, they can only operate on dense ground-truth, which is not available for most depth labels, especially in real-world settings. In this paper we present GRIN, an efficient diffusion model designed to ingest sparse unstructured training data. We use image features with 3D geometric positional encodings to condition the diffusion process both globally and locally, generating depth predictions at a pixel-level. With comprehensive experiments across eight indoor and outdoor datasets, we show that GRIN establishes a new state of the art in zero-shot metric monocular depth estimation even when trained from scratch.

Figures (8)

• Figure 1: GRIN sets a new state of the art in zero-shot metric monocular depth estimation, via efficient pixel-level diffusion and the proper handling of sparse training data. For comparison, we overlay ground-truth metric data with predicted pointclouds.
• Figure 2: Recurrent Interface Networks (RIN) architecture. (a) Latent tokens $\textbf{Z}_{in}$ read from input tokens $\textbf{X}_{in}$, are processed via a series of self-attention layers, and written back to output tokens $\textbf{X}_{out}$. (b) A RIN model consists of $B$ blocks, each receiving latent $\textbf{Z}_{b}$ and input $\textbf{X}_{b}$ tokens from the previous block and returning updated $\textbf{Z}_{b+1}$ and $\textbf{X}_{b+1}$.
• Figure 3: Diagram of GRIN for monocular depth estimation. An input image $\textbf{I}$ with intrinsics $\textbf{K}$ is used to condition the diffusion process both locally, by augmenting each pixel to be predicted with geometrically aware visual features; and globally, by introducing additional scene-level information decoupled from the pixels to be predicted. The resulting tokens are concatenated and attended with the RIN latent space, generating noise predictions for a particular diffusion timestep.
• Figure 4: Qualitative zero-shot metric depth estimation results using GRIN on various indoor and outdoor datasets. The same model was used in all evaluations. For more examples, please refer to the supplementary material.
• Figure 5: Uncertainty estimation analysis using multiple GRIN samples. In (a), Depth and uncertainty maps are calculated taking the median and standard deviation of $s=10$ samples. In (b) we show improvements in depth estimation by only evaluating a percentage of pixels with lower standard deviation. More examples can be found in the supplementary material.