This is the dataset and code release of the OpenRooms Dataset.

Overview

OpenRooms Dataset Release

Zhengqin Li, Ting-Wei Yu, Shen Sang, Sarah Wang, Meng Song, Yuhan Liu, Yu-Ying Yeh, Rui Zhu, Nitesh Gundavarapu, Jia Shi, Sai Bi, Zexiang Xu, Hong-Xing Yu, Kalyan Sunkavalli, Miloš Hašan, Ravi Ramamoorthi, Manmohan Chandraker

Dataset Overview

pipeline

This is the webpage for downloading the OpenRooms dataset. We will first introduce the rendered images and various ground-truths. Later, we will introduce how to render your own images based on the OpenRooms dataset creation pipeline. For each type of data, we offer two kinds of formats, zip files and individual folders, so that users can choose whether to download the whole dataset more efficiently or download individual folders for different scenes. To download the file, we recommend the tool Rclone, otherwise users may suffer from slow downloading speed and instability. If you have any questions, please email to [email protected].

We render six versions of images for all the scenes. Those rendered results are saved in 6 folders: main_xml, main_xml1, mainDiffMat_xml, mainDiffMat_xml1, mainDiffLight_xml and mainDiffLight_xml1. All 6 versions are built with the same CAD models. main_xml, mainDiffMat_xml, mainDiffLight_xml share one set of camera views while main_xml1, mainDiffMat_xml1 and mainDiffLight_xml1 share the other set of camera views. main_xml(1) and mainDiffMat_xml(1) have the same lighting but different materials while main_xml(1) and mainDiffLight_xml(1) have the same materials but different lighting. Both the lighting and material configuration of main_xml and main_xml1 are different. We believe this configuration can potentially help us develope novel applications for image editing. Two example scenes from main_xml, mainDiffMat_xml and mainDiffLight_xml are shown in the below.

config

News: We currently only release the rendered images of the dataset. All ground-truths will be released in a few days. The dataset creation pipeline will also be released soon.

Rendered Images and Ground-truths

All rendered images and the corresponding ground-truths are saved in folder data/rendering/data/. In the following, we will detail each type of rendered data and how to read and interpret them. Two example scenes with images and all ground-truths are included in Demo and Demo.zip.

  1. Images and Images.zip: The 480 × 640 HDR images im_*.hdr, which can be read with the python command.

    im = cv2.imread('im_1.hdr', -1)[:, :, ::-1]

    We render images for main_xml(1), mainDiffMat_xml(1) and mainDiffLight_xml(1).

  2. Material and Material.zip: The 480 × 640 diffuse albedo maps imbaseColor_*.png and roughness map imroughness_*.png. Note that the diffuse albedo map is saved in sRGB space. To load it into linear RGB space, we can use the following python commands. The roughness map is saved in linear space and can be read directly.

    im = cv2.imread('imbaseColor_1.hdr')[:, :, ::-1]
    im = (im.astype(np.float32 ) / 255.0) ** (2.2)

    We only render the diffuse albedo maps and roughness maps for main_xml(1) and mainDiffMat_xml(1) because mainDiffLight_xml(1) share the same material maps with the main_xml(1).

  3. Geometry and Geometry.zip: The 480 × 640 normal maps imnomral_*.png and depth maps imdepth_*.dat. The R, G, B channel of the normal map corresponds to right, up, backward direction of the image plane. To load the depth map, we can use the following python commands.

    with open('imdepth_1.dat', 'rb') as fIn:
        # Read the height and width of depth
        hBuffer = fIn.read(4)
        height = struct.unpack('i', hBuffer)[0]
        wBuffer = fIn.read(4)
        width = struct.unpack('i', wBuffer)[0]
        # Read depth 
        dBuffer = fIn.read(4 * width * height )
        depth = np.array(
            struct.unpack('f' * height * width, dBuffer ), 
            dtype=np.float32 )
        depth = depth.reshape(height, width)

    We render normal maps for main_xml(1) and mainDiffMat_xml(1), and depth maps for main_xml(1).

  4. Mask and Mask.zip: The 480 × 460 grey scale mask immask_*.png for light sources. The pixel value 0 represents the region of environment maps. The pixel value 0.5 represents the region of lamps. Otherwise, the pixel value will be 1. We render the ground-truth masks for main_xml(1) and mainDiffLight_xml(1).

  5. SVLighting: The (120 × 16) × (160 × 32) per-pixel environment maps imenv_*.hdr. The spatial resolution is 120 x 160 while the environment map resolution is 16 x 32. To read the per-pixel environment maps, we can use the following python commands.

    # Read the envmap of resolution 1920 x 5120 x 3 in RGB format 
    env = cv2.imread('imenv_1', -1)[:, :, ::-1]
    # Reshape and permute the per-pixel environment maps
    env = env.reshape(120, 16, 160, 32, 3)
    env = env.transpose(0, 2, 1, 3, 4)

    We render per-pixel environment maps for main_xml(1), mainDiffMat_xml(1) and mainDiffLight_xml(1). Since the total size of per-pixel environment maps is 4.0 TB, we do not provide an extra .zip format for downloading. Please consider using the tool Rclone if you hope to download all the per-pixel environment maps.

  6. SVSG and SVSG.zip: The ground-truth spatially-varying spherical Gaussian (SG) parameters imsgEnv_*.h5, computed from this optimization code. We generate the ground-truth SG parameters for main_xml(1), mainDiffMat_xml(1) and mainDiffLight_xml(1). For the detailed format, please refer to the optimization code.

  7. Shading and Shading.zip: The 120 × 160 diffuse shading imshading_*.hdr computed by intergrating the per-pixel environment maps. We render shading for main_xml(1), mainDiffMat_xml(1) and mainDiffLight_xml(1).

  8. SVLightingDirect and SVLightingDirect.zip: The (30 × 16) × (40 × 32) per-pixel environment maps with direct illumination imenvDirect_*.hdr only. The spatial resolution is 30 × 40 while the environment maps resolution is 16 × 32. The direct per-pixel environment maps can be load the same way as the per-pixel environment maps. We only render direct per-pixel environment maps for main_xml(1) and mainDiffLight_xml(1) because the direct illumination of mainDiffMat_xml(1) is the same as main_xml(1).

  9. ShadingDirect and ShadingDirect.zip: The 120 × 160 direct shading imshadingDirect_*.rgbe. To load the direct shading, we can use the following python command.

    im = cv2.imread('imshadingDirect_1.rgbe', -1)[:, :, ::-1]

    Again, we only render direct shading for main_xml(1) and mainDiffLight_xml(1)

  10. SemanticLabel and SemanticLabel.zip: The 480 × 640 semantic segmentation label imsemLabel_*.npy. We provide semantic labels for 45 classes of commonly seen objects and layout for indoor scenes. The 45 classes can be found in semanticLabels.txt. We only render the semantic labels for main_xml(1).

  11. LightSource and LightSource.zip: The light source information, including geometry, shadow and direct shading of each light source. In each scene directory, light_x directory corresponds to im_x.hdr, where x = 0, 1, 2, 3 ... In each light_x directory, you will see files with numbers in their names. The numbers correspond to the light source ID, i.e. if the IDs are from 0 to 4, then there are 5 light sources in this scene.

    • Geometry: We provide geometry annotation for windows and lamps box_*.dat for main_xml(1) only. To read the annotation, we can use the following python commmands.
      with open('box_0.dat', 'rb')  as fIn:
          info = pickle.load(fIn )
      There are 3 items saved in the dictionary, which we list blow.
      • isWindow: True if the light source is a window, false if the light source is a lamp.
      • box3D: The 3D bounding box of the light source, including center center, orientation xAxis, yAxis, zAxis and size xLen, yLen, zLen.
      • box2D: The 2D bounding box of the light source on the image plane x1, y1, x2, y2.
    • Mask: The 120 × 160 2D binary masks for light sources mask*.png. We only provide the masks for main_xml(1).
    • Direct shading: The 120 × 160 direct shading for each light source imDS*.rgbe. We provide the direction shading for main_xml(1) and mainDiffLight_xml(1).
    • Direct shading without occlusion: The 120 × 160 direct shading with outocclusion for each light source imNoOcclu*.rgbe. We provide the direction shading for main_xml(1) and mainDiffLight_xml(1).
    • Shadow: The 120 × 160 shadow maps for each light source imShadow*.png. We render the shadow map for main_xml(1) only.
  12. Friction and Friction.zip: The friction coefficients computed from our SVBRDF following the method proposed by Zhang et al. We compute the friction coefficients for main_xml(1) and mainDiffLight_xml(1)

Dataset Creation

  1. GPU renderer: The Optix-based GPU path tracer for rendering. Please refer to the github repository for detailed instructions.
  2. Tileable texture synthesis: The tielable texture synthesis code to make sure that the SVBRDF maps are tileable. Please refer to the github repository for more details.
  3. Spherical gaussian optimization: The code to fit per-pixel environment map with spherical Gaussian lobes, using LBFGS optimization. Please refer to the github repository for detailed instructions.

The CAD models, environment maps, materials and code required to recreate the dataset will be released soon.

Applications

  1. Inverse Rendering: Trained on our dataset, we achieved state-of-the-arts on some inverse rendering metrics, especially the lighting estimation. Please refer to our github repository for the training and testing code.
  2. Robotics: Our robotics applications will come soon.

Related Datasets

The OpenRooms dataset is built on the datasets listed below. We thank their creators for the excellent contribution. Please refer to prior datasets for license issues and terms of use if you hope to use them to create your own dataset.

  1. ScanNet dataset: The real 3D scans of indoor scenes.
  2. Scan2cad dataset: The alignment of CAD models to the scanned point clouds.
  3. Laval outdoor lighting dataset: HDR outdoor environment maps
  4. HDRI Haven lighting dataset: HDR outdoor environment maps
  5. PartNet dataset: CAD models
  6. Adobe Stock: High-quality microfacet SVBRDF texture maps. Please license materials from the Adobe website.
Implementation for "Domain-Specific Bias Filtering for Single Labeled Domain Generalization"

DSBF Introduction This repository contains the implementation code for paper: Domain-Specific Bias Filtering for Single Labeled Domain Generalization

ScottYuan 7 Jan 05, 2023
Tutorial on scikit-learn and IPython for parallel machine learning

Parallel Machine Learning with scikit-learn and IPython Video recording of this tutorial given at PyCon in 2013. The tutorial material has been rearra

Olivier Grisel 1.6k Dec 26, 2022
Converts given image (png, jpg, etc) to amogus gif.

Image to Amogus Converter Converts given image (.png, .jpg, etc) to an amogus gif! Usage Place image in the /target/ folder (or anywhere realistically

Hank Magan 1 Nov 24, 2021
A library for differentiable nonlinear optimization.

Theseus A library for differentiable nonlinear optimization built on PyTorch to support constructing various problems in robotics and vision as end-to

Meta Research 1.1k Dec 30, 2022
On Evaluation Metrics for Graph Generative Models

On Evaluation Metrics for Graph Generative Models Authors: Rylee Thompson, Boris Knyazev, Elahe Ghalebi, Jungtaek Kim, Graham Taylor This is the offic

13 Jan 07, 2023
In Search of Probeable Generalization Measures

In Search of Probeable Generalization Measures Exciting News! In Search of Probeable Generalization Measures has been accepted to the International Co

Mahdi S. Hosseini 6 Sep 11, 2022
VSR-Transformer - This paper proposes a new Transformer for video super-resolution (called VSR-Transformer).

VSR-Transformer By Jiezhang Cao, Yawei Li, Kai Zhang, Luc Van Gool This paper proposes a new Transformer for video super-resolution (called VSR-Transf

Jiezhang Cao 225 Nov 13, 2022
🤗 Transformers: State-of-the-art Natural Language Processing for Pytorch, TensorFlow, and JAX.

English | 简体中文 | 繁體中文 State-of-the-art Natural Language Processing for Jax, PyTorch and TensorFlow 🤗 Transformers provides thousands of pretrained mo

Hugging Face 77.2k Jan 02, 2023
A very short and easy implementation of Quantile Regression DQN

Quantile Regression DQN Quantile Regression DQN a Minimal Working Example, Distributional Reinforcement Learning with Quantile Regression (https://arx

Arsenii Senya Ashukha 80 Sep 17, 2022
GemNet model in PyTorch, as proposed in "GemNet: Universal Directional Graph Neural Networks for Molecules" (NeurIPS 2021)

GemNet: Universal Directional Graph Neural Networks for Molecules Reference implementation in PyTorch of the geometric message passing neural network

Data Analytics and Machine Learning Group 124 Dec 30, 2022
Open-source implementation of Google Vizier for hyper parameters tuning

Advisor Introduction Advisor is the hyper parameters tuning system for black box optimization. It is the open-source implementation of Google Vizier w

tobe 1.5k Jan 04, 2023
[ICME 2021 Oral] CORE-Text: Improving Scene Text Detection with Contrastive Relational Reasoning

CORE-Text: Improving Scene Text Detection with Contrastive Relational Reasoning This repository is the official PyTorch implementation of CORE-Text, a

Jingyang Lin 18 Aug 11, 2022
Library of deep learning models and datasets designed to make deep learning more accessible and accelerate ML research.

Tensor2Tensor Tensor2Tensor, or T2T for short, is a library of deep learning models and datasets designed to make deep learning more accessible and ac

12.9k Jan 09, 2023
Open source implementation of "A Self-Supervised Descriptor for Image Copy Detection" (SSCD).

A Self-Supervised Descriptor for Image Copy Detection (SSCD) This is the open-source codebase for "A Self-Supervised Descriptor for Image Copy Detecti

Meta Research 68 Jan 04, 2023
DeepHyper: Scalable Asynchronous Neural Architecture and Hyperparameter Search for Deep Neural Networks

What is DeepHyper? DeepHyper is a software package that uses learning, optimization, and parallel computing to automate the design and development of

DeepHyper Team 214 Jan 08, 2023
Implementation of Pooling by Sliced-Wasserstein Embedding (NeurIPS 2021)

PSWE: Pooling by Sliced-Wasserstein Embedding (NeurIPS 2021) PSWE is a permutation-invariant feature aggregation/pooling method based on sliced-Wasser

Navid Naderializadeh 3 May 06, 2022
3DV 2021: Synergy between 3DMM and 3D Landmarks for Accurate 3D Facial Geometry

SynergyNet 3DV 2021: Synergy between 3DMM and 3D Landmarks for Accurate 3D Facial Geometry Cho-Ying Wu, Qiangeng Xu, Ulrich Neumann, CGIT Lab at Unive

Cho-Ying Wu 239 Jan 06, 2023
A PyTorch implementation of "Semi-Supervised Graph Classification: A Hierarchical Graph Perspective" (WWW 2019)

SEAL ⠀⠀⠀ A PyTorch implementation of Semi-Supervised Graph Classification: A Hierarchical Graph Perspective (WWW 2019) Abstract Node classification an

Benedek Rozemberczki 202 Dec 27, 2022
Python script for performing depth completion from sparse depth and rgb images using the msg_chn_wacv20. model in ONNX

ONNX msg_chn_wacv20 depth completion Python script for performing depth completion from sparse depth and rgb images using the msg_chn_wacv20 model in

Ibai Gorordo 19 Oct 22, 2022
Official PyTorch implementation of "AASIST: Audio Anti-Spoofing using Integrated Spectro-Temporal Graph Attention Networks"

AASIST This repository provides the overall framework for training and evaluating audio anti-spoofing systems proposed in 'AASIST: Audio Anti-Spoofing

Clova AI Research 56 Jan 02, 2023