Skip to content


Repository files navigation

SimpleRecon: 3D Reconstruction Without 3D Convolutions

This is the reference PyTorch implementation for training and testing MVS depth estimation models using the method described in

SimpleRecon: 3D Reconstruction Without 3D Convolutions

Mohamed Sayed, John Gibson, Jamie Watson, Victor Adrian Prisacariu, Michael Firman, and Clément Godard

Paper, ECCV 2022 (arXiv pdf), Supplemental Material, Project Page, Video

example output


This code is for non-commercial use; please see the license file for terms. If you do find any part of this codebase helpful, please cite our paper using the BibTex below and link this repo. Thanks!

🆕 Updates

25/05/2023: Fixed package verions for llvm-openmp, clang, and protobuf. Do use this new environment file if you have trouble running the code and/or if dataloading is being limited to a single thread.

09/03/2023: Added kornia version to the environments file to fix kornia typing issue. (thanks @natesimon!)

26/01/2023: The license has been modified to make running the model for academic reasons easier. Please the LICENSE file for the exact details.

There is an update as of 31/12/2022 that fixes slightly wrong intrinsics, flip augmentation for the cost volume, and a numerical precision bug in projection. All scores improve. You will need to update your forks and use new weights. See Bug Fixes.

Precomputed scans for online default frames are here:

Table of Contents

🗺️ Overview

SimpleRecon takes as input posed RGB images, and outputs a depth map for a target image.

⚙️ Setup

Assuming a fresh Anaconda distribution, you can install dependencies with:

conda env create -f simplerecon_env.yml

We ran our experiments with PyTorch 1.10, CUDA 11.3, Python 3.9.7 and Debian GNU/Linux 10.

📦 Models

Download a pretrained model into the weights/ folder.

We provide the following models (scores are with online default keyframes):

--config Model Abs Diff↓ Sq Rel↓ delta < 1.05↑ Chamfer↓ F-Score↑
hero_model.yaml Metadata + Resnet Matching 0.0868 0.0127 74.26 5.69 0.680
dot_product_model.yaml Dot Product + Resnet Matching 0.0910 0.0134 71.90 5.92 0.667

hero_model is the one we use in the paper as Ours

🚀 Speed

--config Model Inference Speed (--batch_size 1) Inference GPU memory Approximate training time
hero_model Hero, Metadata + Resnet 130ms / 70ms (speed optimized) 2.6GB / 5.7GB (speed optimized) 36 hours
dot_product_model Dot Product + Resnet 80ms 2.6GB 36 hours

With larger batches speed increases considerably. With batch size 8 on the non-speed optimized model, the latency drops to ~40ms.

📝 TODOs:

  • Simple scan for folks to quickly try the code, instead of downloading the ScanNetv2 test scenes. DONE
  • ScanNetv2 extraction, ETA 10th October DONE
  • FPN model weights.
  • [ ] Tutorial on how to use Scanniverse data, ETA 5th October 10th October 20th October At present there is no publically available way of exporting scans from Scanniverse. You'll have to use ios-logger; NeuralRecon have a good tutorial on this, and a dataloader that accepts the processed format is at datasets/ UPDATE: There is now a quick readme data_scripts/ for how to process and run inference an ios-logger scan using the script at data_scripts/

🏃 Running out of the box!

We've now included two scans for people to try out immediately with the code. You can download these scans from here.


  1. Download weights for the hero_model into the weights directory.
  2. Download the scans and unzip them to a directory of your choosing.
  3. Modify the value for the option dataset_path in configs/data/vdr_dense.yaml to the base path of the unzipped vdr folder.
  4. You should be able to run it! Something like this will work:
            --output_base_path OUTPUT_PATH \
            --config_file configs/models/hero_model.yaml \
            --load_weights_from_checkpoint weights/hero_model.ckpt \
            --data_config configs/data/vdr_dense.yaml \
            --num_workers 8 \
            --batch_size 2 \
            --fast_cost_volume \
            --run_fusion \
            --depth_fuser open3d \
            --fuse_color \

This will output meshes, quick depth viz, and socres when benchmarked against LiDAR depth under OUTPUT_PATH.

This command uses vdr_dense.yaml which will generate depths for every frame and fuse them into a mesh. In the paper we report scores with fused keyframes instead, and you can run those using vdr_default.yaml. You can also use dense_offline tuples by instead using vdr_dense_offline.yaml.

See the section below on testing and evaluation. Make sure to use the correct config flags for datasets.

💾 ScanNetv2 Dataset

Please follow the instructions here to download the dataset. This dataset is quite big (>2TB), so make sure you have enough space, especially for extracting files.

Once downloaded, use this script to export raw sensor data to images and depth files.

We've written a quick tutorial and included modified scripts to help you with downloading and extracting ScanNetv2. You can find them at data_scripts/scannet_wrangling_scripts/

You should change the dataset_path config argument for ScanNetv2 data configs at configs/data/ to match where your dataset is.

The codebase expects ScanNetv2 to be in the following format:

    scans_test (test scans)
            scene0707_00_vh_clean_2.ply (gt mesh)
                frame-000261.color.512.png (optional, image at 512x384)
                frame-000261.color.640.png (optional, image at 640x480)
                frame-000261.depth.png (full res depth, stored scale *1000)
                frame-000261.depth.256.png (optional, depth at 256x192 also
            scene0707.txt (scan metadata and image sizes)
    scans (val and train scans)
            (see above)

In this example scene0707.txt should contain the scan's metadata:

    colorHeight = 968
    colorToDepthExtrinsics = 0.999263 -0.010031 0.037048 ........
    colorWidth = 1296
    depthHeight = 480
    depthWidth = 640
    fx_color = 1170.187988
    fx_depth = 570.924255
    fy_color = 1170.187988
    fy_depth = 570.924316
    mx_color = 647.750000
    mx_depth = 319.500000
    my_color = 483.750000
    my_depth = 239.500000
    numColorFrames = 784
    numDepthFrames = 784
    numIMUmeasurements = 1632

frame-000261.pose.txt should contain pose in the form:

    -0.384739 0.271466 -0.882203 4.98152
    0.921157 0.0521417 -0.385682 1.46821
    -0.0587002 -0.961035 -0.270124 1.51837

frame-000261.color.512.png and frame-000261.color.640.png are precached resized versions of the original image to save load and compute time during training and testing. frame-000261.depth.256.png is also a precached resized version of the depth map.

All resized precached versions of depth and images are nice to have but not required. If they don't exist, the full resolution versions will be loaded, and downsampled on the fly.

🖼️🖼️🖼️ Frame Tuples

By default, we estimate a depth map for each keyframe in a scan. We use DeepVideoMVS's heuristic for keyframe separation and construct tuples to match. We use the depth maps at these keyframes for depth fusion. For each keyframe, we associate a list of source frames that will be used to build the cost volume. We also use dense tuples, where we predict a depth map for each frame in the data, and not just at specific keyframes; these are mostly used for visualization.

We generate and export a list of tuples across all scans that act as the dataset's elements. We've precomputed these lists and they are available at data_splits under each dataset's split. For ScanNet's test scans they are at data_splits/ScanNetv2/standard_split. Our core depth numbers are computed using data_splits/ScanNetv2/standard_split/test_eight_view_deepvmvs.txt.

Here's a quick taxonamy of the type of tuples for test:

  • default: a tuple for every keyframe following DeepVideoMVS where all source frames are in the past. Used for all depth and mesh evaluation unless stated otherwise. For ScanNet use data_splits/ScanNetv2/standard_split/test_eight_view_deepvmvs.txt.
  • offline: a tuple for every frame in the scan where source frames can be both in the past and future relative to the current frame. These are useful when a scene is captured offline, and you want the best accuracy possible. With online tuples, the cost volume will contain empty regions as the camera moves away and all source frames lag behind; however with offline tuples, the cost volume is full on both ends, leading to a better scale (and metric) estimate.
  • dense: an online tuple (like default) for every frame in the scan where all source frames are in the past. For ScanNet this would be data_splits/ScanNetv2/standard_split/test_eight_view_deepvmvs_dense.txt.
  • offline: an offline tuple for every keyframefor every keyframe in the scan.

For the train and validation sets, we follow the same tuple augmentation strategy as in DeepVideoMVS and use the same core generation script.

If you'd like to generate these tuples yourself, you can use the scripts at data_scripts/ for train tuples and data_scripts/ for test tuples. These follow the same config format as and will use whatever dataset class you build to read pose informaiton.

Example for test:

# default tuples
python ./data_scripts/ 
    --data_config configs/data/scannet_default_test.yaml
    --num_workers 16

# dense tuples
python ./data_scripts/ 
    --data_config configs/data/scannet_dense_test.yaml
    --num_workers 16

Examples for train:

# train
python ./data_scripts/ 
    --data_config configs/data/scannet_default_train.yaml
    --num_workers 16

# val
python ./data_scripts/ 
    --data_config configs/data/scannet_default_val.yaml
    --num_workers 16

These scripts will first check each frame in the dataset to make sure it has an existing RGB frame, an existing depth frame (if appropriate for the dataset), and also an existing and valid pose file. It will save these valid_frames in a text file in each scan's folder, but if the directory is read only, it will ignore saving a valid_frames file and generate tuples anyway.

📊 Testing and Evaluation

You can use for inferring and evaluating depth maps and fusing meshes.

All results will be stored at a base results folder (results_path) at:


where opts is the options class. For example, when opts.output_base_path is ./results, is HERO_MODEL, opts.dataset is scannet, and opts.frame_tuple_type is default, the output directory will be


Make sure to set --opts.output_base_path to a directory suitable for you to store results.

--frame_tuple_type is the type of image tuple used for MVS. A selection should be provided in the data_config file you used.

By default will attempt to compute depth scores for each frame and provide both frame averaged and scene averaged metrics. The script will save these scores (per scene and totals) under results_path/scores.

We've done our best to ensure that a torch batching bug through the matching encoder is fixed for (<10^-4) accurate testing by disabling image batching through that encoder. Run --batch_size 4 at most if in doubt, and if you're looking to get as stable as possible numbers and avoid PyTorch gremlins, use --batch_size 1 for comparison evaluation.

If you want to use this for speed, set --fast_cost_volume to True. This will enable batching through the matching encoder and will enable an einops optimized feature volume.

# Example command to just compute scores 
            --output_base_path OUTPUT_PATH \
            --config_file configs/models/hero_model.yaml \
            --load_weights_from_checkpoint weights/hero_model.ckpt \
            --data_config configs/data/scannet_default_test.yaml \
            --num_workers 8 \
            --batch_size 4;

# If you'd like to get a super fast version use:
            --output_base_path OUTPUT_PATH \
            --config_file configs/models/hero_model.yaml \
            --load_weights_from_checkpoint weights/hero_model.ckpt \
            --data_config configs/data/scannet_default_test.yaml \
            --num_workers 8 \
            --fast_cost_volume \
            --batch_size 2;

This script can also be used to perform a few different auxiliary tasks, including:

TSDF Fusion

To run TSDF fusion provide the --run_fusion flag. You have two choices for fusers

  1. --depth_fuser ours (default) will use our fuser, whose meshes are used in most visualizations and for scores. This fuser does not support color. We've provided a custom branch of scikit-image with our custom implementation of measure.matching_cubes that allows single walled. We use single walled meshes for evaluation. If this is isn't important to you, you can set the export_single_mesh to False for call to export_mesh in
  2. --depth_fuser open3d will use the open3d depth fuser. This fuser supports color and you can enable this by using the --fuse_color flag.

By default, depth maps will be clipped to 3m for fusion and a tsdf resolution of 0.04m3 will be used, but you can change that by changing both --max_fusion_depth and --fusion_resolution

You can optionnally ask for predicted depths used for fusion to be masked when no vaiid MVS information exists using --mask_pred_depths. This is not enabled by default.

You can also fuse the best guess depths from the cost volume before the cost volume encoder-decoder that introduces a strong image prior. You can do this by using --fusion_use_raw_lowest_cost.

Meshes will be stored under results_path/meshes/.

# Example command to fuse depths to get meshes
            --output_base_path OUTPUT_PATH \
            --config_file configs/models/hero_model.yaml \
            --load_weights_from_checkpoint weights/hero_model.ckpt \
            --data_config configs/data/scannet_default_test.yaml \
            --num_workers 8 \
            --run_fusion \
            --batch_size 8;

Cache depths

You can optionally store depths by providing the --cache_depths flag. They will be stored at results_path/depths.

# Example command to compute scores and cache depths
            --output_base_path OUTPUT_PATH \
            --config_file configs/models/hero_model.yaml \
            --load_weights_from_checkpoint weights/hero_model.ckpt \
            --data_config configs/data/scannet_default_test.yaml \
            --num_workers 8 \
            --cache_depths \
            --batch_size 8;

# Example command to fuse depths to get color meshes
            --output_base_path OUTPUT_PATH \
            --config_file configs/models/hero_model.yaml \
            --load_weights_from_checkpoint weights/hero_model.ckpt \
            --data_config configs/data/scannet_default_test.yaml \
            --num_workers 8 \
            --run_fusion \
            --depth_fuser open3d \
            --fuse_color \
            --batch_size 4;

Quick viz

There are other scripts for deeper visualizations of output depths and fusion, but for quick export of depth map visualization you can use --dump_depth_visualization. Visualizations will be stored at results_path/viz/quick_viz/.

# Example command to output quick depth visualizations
            --output_base_path OUTPUT_PATH \
            --config_file configs/models/hero_model.yaml \
            --load_weights_from_checkpoint weights/hero_model.ckpt \
            --data_config configs/data/scannet_default_test.yaml \
            --num_workers 8 \
            --dump_depth_visualization \
            --batch_size 4;

👉☁️ Point Cloud Fusion

We also allow point cloud fusion of depth maps using the fuser from 3DVNet's repo.

# Example command to fuse depths into point clouds.
            --output_base_path OUTPUT_PATH \
            --config_file configs/models/hero_model.yaml \
            --load_weights_from_checkpoint weights/hero_model.ckpt \
            --data_config configs/data/scannet_dense_test.yaml \
            --num_workers 8 \
            --batch_size 4;

Change configs/data/scannet_dense_test.yaml to configs/data/scannet_default_test.yaml to use keyframes only if you don't want to wait too long.

📊 Mesh Metrics

We use TransformerFusion's mesh evaluation for our main results table but set the seed to a fixed value for consistency when randomly sampling meshes. We also report mesh metrics using NeuralRecon's evaluation in the supplemental material.

For point cloud evaluation, we use TransformerFusion's code but load in a point cloud in place of sampling a mesh's surface.

⏳ Training

By default models and tensorboard event files are saved to ~/tmp/tensorboard/<model_name>. This can be changed with the --log_dir flag.

We train with a batch_size of 16 with 16-bit precision on two A100s on the default ScanNetv2 split.

Example command to train with two GPUs:

            --log_dir logs \
            --config_file configs/models/hero_model.yaml \
            --data_config configs/data/scannet_default_train.yaml \
            --gpus 2 \
            --batch_size 16;

The code supports any number of GPUs for training. You can specify which GPUs to use with the CUDA_VISIBLE_DEVICES environment.

All our training runs were performed on two NVIDIA A100s.

Different dataset

You can train on a custom MVS dataset by writing a new dataloader class which inherits from GenericMVSDataset at datasets/ See the ScannetDataset class in datasets/ or indeed any other class in datasets for an example.

🎛️ Finetuning a pretrained model

To finetune, simple load a checkpoint (not resume!) and train from there:

CUDA_VISIBLE_DEVICES=0 python --config configs/models/hero_model.yaml
                --data_config configs/data/scannet_default_train.yaml 
                --load_weights_from_checkpoint weights/hero_model.ckpt

Change the data configs to whatever dataset you want to finetune to.

🔧 Other training and testing options

See for the range of other training options, such as learning rates and ablation settings, and testing options.

✨ Visualization

Other than quick depth visualization in the script, there are two scripts for visualizing depth output.

The first is visualization_scripts/ This will produce a video with color images of the reference and source frames, depth prediction, cost volume estimate, GT depth, and estimated normals from depth. The script assumes you have cached depth output using and accepts the same command template format as

# Example command to get visualizations for dense frames
CUDA_VISIBLE_DEVICES=0 python ./visualization_scripts/ --name HERO_MODEL \
            --output_base_path OUTPUT_PATH \
            --data_config configs/data/scannet_dense_test.yaml \
            --num_workers 8;

where OUTPUT_PATH is the base results directory for SimpleRecon (what you used for test to begin with). You could optionally run .visualization_scripts/ before this script to get a scene average for the min and max depth values used for colormapping; if those aren't available, the script will use 0m and 5m for colomapping min and max.

The second allows a live visualization of meshing. This script will use cached depth maps if available, otherwise it will use the model to predict them before fusion. The script will iteratively load in a depth map, fuse it, save a mesh file at this step, and render this mesh alongside a camera marker for the birdseye video, and from the point of view of the camera for the fpv video.

# Example command to get live visualizations for mesh reconstruction
            --output_base_path OUTPUT_PATH \
            --config_file configs/models/hero_model.yaml \
            --load_weights_from_checkpoint weights/hero_model.ckpt \
            --data_config configs/data/scannet_dense_test.yaml \
            --num_workers 8;

By default the script will save meshes to an intermediate location, and you can optionally load those meshes to save time when visualizing the same meshes again by passing --use_precomputed_partial_meshes. All intermediate meshes will have had to be computed on the previous run for this to work.

📝🧮👩‍💻 Notation for Transformation Matrices

TL;DR: world_T_cam == world_from_cam
This repo uses the notation "cam_T_world" to denote a transformation from world to camera points (extrinsics). The intention is to make it so that the coordinate frame names would match on either side of the variable when used in multiplication from right to left:

cam_points = cam_T_world @ world_points

world_T_cam denotes camera pose (from cam to world coords). ref_T_src denotes a transformation from a source to a reference view.
Finally this notation allows for representing both rotations and translations such as: world_R_cam and world_t_cam

🗺️ World Coordinate System

This repo is geared towards ScanNet, so while its functionality should allow for any coordinate system (signaled via input flags), the model weights we provide assume a ScanNet coordinate system. This is important since we include ray information as part of metadata. Other datasets used with these weights should be transformed to the ScanNet system. The dataset classes we include will perform the appropriate transforms.

🐜🔧 Bug Fixes

Update 31/12/2022:

There are a few bugs addressed in this update, you will need to update your forks and use new weights from the table near the beginning of this README. You will also need to make sure you have the correct intrinsics files extracted using the reader.

  • We were initially using a slightly incorrect set of intrinsics in ScanNet. The repo now uses intriniscs from the intriniscs folder.
  • The MLP in the cost volume wasn't seeing any flip augmentation which led to biases around edges, so we've now included a geometry based flip in the base dataset class. It is enabled only for the train split.
  • We had a bug in projection that never allowed the mask in the cost volume to properly function, so we've now switched to using the same normalization as in OpenCV and Kornia.

Thanks to all those that pointed it out and were patient while we worked on fixes.

All scores improve with these fixes, and the associated weights are uploaded here. For old scores, code, and weights, check this commit hash: 7de5b451e340f9a11c7fd67bd0c42204d0b009a9

Full scores for models with bug fixes:


--config Abs Diff↓ Abs Rel↓ Sq Rel↓ RMSE↓ log RMSE↓ delta < 1.05↑ delta < 1.10↑
hero_model.yaml, Metadata + Resnet 0.0868 0.0428 0.0127 0.1472 0.0681 74.26 90.88
dot_product_model.yaml, dot product + Resnet 0.0910 0.0453 0.0134 0.1509 0.0704 71.90 89.75

Mesh Fusion

--config Acc↓ Comp↓ Chamfer↓ Recall↑ Precision↑ F-Score↑
hero_model.yaml, Metadata + Resnet 5.41 5.98 5.69 0.695 0.668 0.680
dot_product_model.yaml, dot product + Resnet 5.66 6.18 5.92 0.682 0.655 0.667


--config Model Abs Diff↓ Sq Rel↓ delta < 1.05↑ Chamfer↓ F-Score↑
hero_model.yaml Metadata + Resnet Matching 0.0868 0.0127 74.26 5.69 0.680
OLD hero_model.yaml Metadata + Resnet Matching 0.0885 0.0125 73.16 5.81 0.671
dot_product_model.yaml Dot Product + Resnet Matching 0.0910 0.0134 71.90 5.92 0.667
OLD dot_product_model.yaml Dot Product + Resnet Matching 0.0941 0.0139 70.48 6.29 0.642

Tiny bug with frame count:

Initially this repo spat out tuple files for default DVMVS style keyframes with 9 extra frame of 25599 for the ScanNetv2 test set. There was a minor bug with handling lost tracking that's now fixed. This repo should now mimic the DVMVS keyframe buffer exactly, with 25590 keyframes for testing. The only effect this bug had was the inclusion of 9 extra frames, all the other tuples were exactly the same as that of DVMVS. The offending frames are in these scans

scan         previous count  new count
scene0711_00 393             392
scene0727_00 209             208 
scene0736_00 1023            1022 
scene0737_00 408             407 
scene0751_00 165             164 
scene0775_00 220             219 
scene0791_00 227             226 
scene0794_00 141             140 
scene0795_00 102             101 

The tuple files for default test have been updated. Since this is a small (~3e-4) difference in extra frames scored, the scores are unchanged.

🗺️💾 COLMAP Dataset

TL;DR: Scale your poses and crop your images.

We do provide a dataloader for loading images from a COLMAP sparse reconstruction. For this to work with SimpleRecon, you'll need to crop your images to match the FOV of ScanNet (roughly similar to an iPhone's FOV in video mode), and scale your pose's location using known real world measurements. If these steps aren't taken, the cost volume won't be built correctly, and the network will not estimate depth properly.

🙏 Acknowledgements

We thank Aljaž Božič of TransformerFusion, Jiaming Sun of Neural Recon, and Arda Düzçeker of DeepVideoMVS for quickly providing useful information to help with baselines and for making their codebases readily available, especially on short notice.

The tuple generation scripts make heavy use of a modified version of DeepVideoMVS's Keyframe buffer (thanks again Arda and co!).

The PyTorch point cloud fusion module at torch_point_cloud_fusion code is borrowed from 3DVNet's repo. Thanks Alexander Rich!

We'd also like to thank Niantic's infrastructure team for quick actions when we needed them. Thanks folks!

Mohamed is funded by a Microsoft Research PhD Scholarship (MRL 2018-085).

📜 BibTeX

If you find our work useful in your research please consider citing our paper:

  title={SimpleRecon: 3D Reconstruction Without 3D Convolutions},
  author={Sayed, Mohamed and Gibson, John and Watson, Jamie and Prisacariu, Victor and Firman, Michael and Godard, Cl{\'e}ment},
  booktitle={Proceedings of the European Conference on Computer Vision (ECCV)},

👩‍⚖️ License

Copyright © Niantic, Inc. 2022. Patent Pending. All rights reserved. Please see the license file for terms.