(CVPR 2022) TransMVSNet: Global Context-aware Multi-view Stereo Network with Transformers.
Tips: If you meet any problems when reproduce our results, please contact Yikang Ding ([email protected]). We are happy to help you solve the problems and share our experience.
In this paper, we present TransMVSNet, based on our exploration of feature matching in multi-view stereo (MVS). We analogize MVS back to its nature of a feature matching task and therefore propose a powerful Feature Matching Transformer (FMT) to leverage intra- (self-) and inter- (cross-) attention to aggregate long-range context information within and across images. To facilitate a better adaptation of the FMT, we leverage an Adaptive Receptive Field (ARF) module to ensure a smooth transit in scopes of features and bridge different stages with a feature pathway to pass transformed features and gradients across different scales. In addition, we apply pair-wise feature correlation to measure similarity between features, and adopt ambiguity-reducing focal loss to strengthen the supervision. To the best of our knowledge, TransMVSNet is the first attempt to leverage Transformer into the task of MVS. As a result, our method achieves state-of-the-art performance on DTU dataset, Tanks and Temples benchmark, and BlendedMVS dataset.
Our code is tested with Python==3.6/3.7/3.8, PyTorch==1.6.0/1.7.0/1.9.0, CUDA==10.2 on Ubuntu-18.04 with NVIDIA GeForce RTX 2080Ti. Similar or higher version should work well.
To use TransMVSNet, clone this repo:
git clone https://github.com/MegviiRobot/TransMVSNet.git
cd TransMVSNet
We highly recommend using Anaconda to manage the python environment:
conda create -n transmvsnet python=3.6
conda activate transmvsnet
pip install -r requirements.txt
We also recommend using apex, you can install apex from the official repo.
In TransMVSNet, we mainly use DTU, BlendedMVS and Tanks and Temples to train and evaluate our models. You can prepare the corresponding data by following the instructions below.
For DTU training set, you can download the preprocessed DTU training data and Depths_raw (both from Original MVSNet), and unzip them to construct a dataset folder like:
dtu_training
βββ Cameras
βββ Depths
βββ Depths_raw
βββ Rectified
For DTU testing set, you can download the preprocessed DTU testing data (from Original MVSNet) and unzip it as the test data folder, which should contain one cams
folder, one images
folder and one pair.txt
file.
We use the low-res set of BlendedMVS dataset for both training and testing. You can download the low-res set from orignal BlendedMVS and unzip it to form the dataset folder like below:
BlendedMVS
βββ 5a0271884e62597cdee0d0eb
β βββ blended_images
β βββ cams
β βββ rendered_depth_maps
βββ 59338e76772c3e6384afbb15
βββ 59f363a8b45be22330016cad
βββ ...
βββ all_list.txt
βββ training_list.txt
βββ validation_list.txt
Download our preprocessed Tanks and Temples dataset and unzip it to form the dataset folder like below:
tankandtemples
βββ advanced
β βββ Auditorium
β βββ Ballroom
β βββ ...
β βββ Temple
βββ intermediate
βββ Family
βββ Francis
βββ ...
βββ Train
Set the configuration in scripts/train.sh
:
MVS_TRAINING
as the path of DTU training set.LOG_DIR
to save the checkpoints.NGPUS
to suit your device.torch.distributed.launch
by default.To train your own model, just run:
bash scripts/train.sh
You can conveniently modify more hyper-parameters in scripts/train.sh
according to the argparser in train.py
, such as summary_freq
, save_freq
, and so on.
For a fair comparison with other SOTA methods on Tanks and Temples benchmark, we finetune our model on BlendedMVS dataset after training on DTU dataset.
Set the configuration in scripts/train_bld_fintune.sh
:
MVS_TRAINING
as the path of BlendedMVS dataset.LOG_DIR
to save the checkpoints and training log.CKPT
as path of the loaded .ckpt
which is trained on DTU dataset.To finetune your own model, just run:
bash scripts/train_bld_fintune.sh
For easy testing, you can download our pre-trained models and put them in checkpoints
folder, or use your own models and follow the instruction below.
Important Tips: to reproduce our reported results, you need to:
gipuma
from Yao Yao as introduced belowalign_corner
in several functions, which would affect the final resultsmodel_dtu.ckpt
for testingTo start testing, set the configuration in scripts/test_dtu.sh
:
TESTPATH
as the path of DTU testing set.TESTLIST
as the path of test list (.txt file).CKPT_FILE
as the path of the model weights.OUTDIR
as the path to save results.Run:
bash scripts/test_dtu.sh
Note: You can use the gipuma
fusion method or normal
fusion method to fuse the point clouds. In our experiments, we use the gipuma
fusion method by default.
With using the uploaded ckpt and latest code, these two fusion methods would get the below results:
Fuse | Overall |
---|---|
gipuma | 0.304 |
normal | 0.314 |
To install the gipuma
, clone the modified version from Yao Yao.
Modify the line-10 in CMakeLists.txt
to suit your GPUs. Othervise you would meet warnings when compile it, which would lead to failure and get 0 points in fused point cloud. For example, if you use 2080Ti GPU, modify the line-10 to:
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS};-O3 --use_fast_math --ptxas-options=-v -std=c++11 --compiler-options -Wall -gencode arch=compute_70,code=sm_70)
If you use other kind of GPUs, please modify the arch code to suit your device (arch=compute_XX,code=sm_XX
).
Then install it by cmake .
and make
, which will generate the executable file at FUSIBILE_EXE_PATH
.
Please note
For quantitative evaluation on DTU dataset, download SampleSet and Points. Unzip them and place Points
folder in SampleSet/MVS Data/
. The structure looks like:
SampleSet
βββMVS Data
βββPoints
In DTU-MATLAB/BaseEvalMain_web.m
, set dataPath
as path to SampleSet/MVS Data/
, plyPath
as directory that stores the reconstructed point clouds and resultsPath
as directory to store the evaluation results. Then run DTU-MATLAB/BaseEvalMain_web.m
in matlab.
We also upload our final point cloud results to here. You can easily download them and evaluate them using the MATLAB
scripts, the results look like:
Acc. (mm) | Comp. (mm) | Overall (mm) |
---|---|---|
0.321 | 0.289 | 0.305 |
We recommend using the finetuned models (model_bld.ckpt
) to test on Tanks and Temples benchmark.
Similarly, set the configuration in scripts/test_tnt.sh
:
TESTPATH
as the path of intermediate set or advanced set.TESTLIST
as the path of test list (.txt file).CKPT_FILE
as the path of the model weights.OUTDIR
as the path to save resutls.To generate point cloud results, just run:
bash scripts/test_tnt.sh
Note thatοΌ
For quantitative evaluation, you can upload your point clouds to Tanks and Temples benchmark.
@inproceedings{ding2022transmvsnet,
title={Transmvsnet: Global context-aware multi-view stereo network with transformers},
author={Ding, Yikang and Yuan, Wentao and Zhu, Qingtian and Zhang, Haotian and Liu, Xiangyue and Wang, Yuanjiang and Liu, Xiao},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={8585--8594},
year={2022}
}
We borrow some code from CasMVSNet, LoFTR and AA-RMVSNet. We thank the authors for releasing the source code.