[Paper Review] DIODE: A Dense Indoor and Outdoor DEpth Dataset

Igor Vasiljevic, Nicholas Kolkin+9 more|arXiv (Cornell University)|Aug 1, 2019

Advanced Vision and Imaging32 references52 citations

TL;DR

DIODE provides a large, high-resolution RGB-D dataset with dense, accurate depth captured in both indoor and outdoor scenes using a single sensor suite (FARO Focus S350), enabling unified depth estimation across domains.

ABSTRACT

We introduce DIODE, a dataset that contains thousands of diverse high resolution color images with accurate, dense, long-range depth measurements. DIODE (Dense Indoor/Outdoor DEpth) is the first public dataset to include RGBD images of indoor and outdoor scenes obtained with one sensor suite. This is in contrast to existing datasets that focus on just one domain/scene type and employ different sensors, making generalization across domains difficult. The dataset is available for download at http://diode-dataset.org

Motivation & Objective

• Motivate the need for a large, diverse RGB-D dataset that spans indoor and outdoor environments using a single sensing modality.
• Provide dense, accurate depth maps and surface normals to enable high-quality depth estimation and 3D reasoning.
• Establish standardized train/validation/test splits to enable reproducible evaluation and cross-domain generalization.

Proposed method

• Use a FARO Focus S350 laser scanner to collect dense depth over 0.6–350 m with ~1 mm precision and near-1° angular resolution in both indoor and outdoor scenes.
• Rectify scans into 768×1024 RGB crops across multiple viewing frustums and compute robust depth maps and normals per crop via ray-based mapping and RANSAC-based plane fitting.
• Create automated validity masks to filter spurious depth returns and manually mask problematic regions in the validation set.
• Provide a standard train/validation/test split with non-overlapping viewpoints to ensure reproducible evaluation.

Experimental results

Research questions

• RQ1 Can a single high-density RGB-D sensor setup yield accurate, dense depth maps for both indoor and outdoor scenes?
• RQ2 How does monocular depth estimation perform on a unified indoor/outdoor dataset compared to domain-specific RGB-D datasets?
• RQ3 What is the impact of dataset scale and diversity on depth estimation accuracy across indoor and outdoor environments?

Key findings

• DIODE offers high-density depth with up to 350 m range and ~1 mm precision, covering both indoor and outdoor scenes with unified sensing.
• Baseline monocular depth estimation on DIODE (DenseDepth architecture) shows varying transferability between indoor and outdoor subsets, with overall better performance when trained on the full indoor+outdoor dataset.
• Depth distributions on DIODE are more diverse and span a broader range than KITTI or Make3D, highlighting the need for cross-domain depth models.
• Rectified crops and depth maps enable higher-resolution evaluation of depth perception tasks compared to prior indoor- or outdoor-only datasets.