Need help with stereo-transformer?
Click the “chat” button below for chat support from the developer who created it, or find similar developers for support.

About the developer

329 Stars 54 Forks Apache License 2.0 26 Commits 14 Opened issues


Official Repo for Stereo Transformer: Revisiting Stereo Depth Estimation From a Sequence-to-Sequence Perspective with Transformers. (ICCV 2021 Oral)

Services available


Need anything else?

Contributors list

# 157,439
20 commits

STereo TRansformer (STTR)

This is the official repo for our work Revisiting Stereo Depth Estimation From a Sequence-to-Sequence Perspective with Transformers.

Fine-tuned result on street scene:

Generalization to medical domain when trained only on synthetic data:

If you find our work relevant, please cite

  title={Revisiting Stereo Depth Estimation From a Sequence-to-Sequence Perspective with Transformers},
  author={Li, Zhaoshuo and Liu, Xingtong and Drenkow, Nathan and Ding, Andy and Creighton, Francis X and Taylor, Russell H and Unberath, Mathias},
  journal={arXiv preprint arXiv:2011.02910},


  • 2021.03.29: Added code/instruction to obtain training data from Scene Flow.
  • 2021.01.13: STTR-light is released. Use branch
    for the new model.
  • 2020.11.05: First code and arxiv release

Table of Content


Benefits of STTR

STereo TRansformer (STTR) revisits stereo depth estimation from a sequence-to-sequence perspective. The network combines conventional CNN feature extractor and long-range relationship capturing module Transformer. STTR is able to relax prior stereo depth estimation networks in three aspects:

  • Disparity range naturally scales with image resolution, no more manually set range.
  • Explicit occlusion handling.
  • Imposing uniqueness constraint.

STTR performs comparably well against prior work with refinement in Scene Flow and KITTI 2015. STTR is also able to generalize to MPI Sintel, KITTI 2015, Middlebury 2014 and SCARED when trained only on synthetic data.

Working Theory


Two types of attention mechanism are used: self-attention and cross-attention. Self-attention uses context within the same image, while cross-attention uses context across two images. The attention shrinks from global context to local context as the layer goes deeper. Attention in a large textureless area tends to keep attending dominant features like edges, which helps STTR to resolve ambiguity.

Self-Attention Self-attention

Cross-Attention Cross-attention

Relative Positional Encoding

We find that only image-image based attention is not enough. Therefore, we opt in relative positional encoding to provide positional information. This allows STTR to use the relative distance from a featureless pixel to dominant pixel (such as edge) to resolve ambiguity. In the following example, STTR starts to texture the center of the table using relative distance, thus strides parallel to the edges start to show.

Feature Descriptor Feature Descriptor

Implicit Learnt Feature Classification

We observe that the feature extractor before Transformer actually learns without any explicit supervision to classify pixels into two clusters - textured and textureless. We hypothesize that this implicit learnt classification helps STTR to generalize.

Implicit Learnt Classification Implicit Learnt Classification


We recommend the following steps to set up your environment - Create your python virtual environment by

    conda create --name sttr python=3.6 # create a virtual environment called "sttr" with python version 3.6
(as long as it is Python 3, it can be anything >= 3.6) - Install Pytorch 1.5.1: Please follow link here. - Other versions of Pytorch may have problem during evaluation, see Issue #8 for more detail.
  • Other third-party packages: You can use pip to install the dependencies by
    pip install -r requirements.txt
  • (Optional) Install Nvidia apex: We use apex for mixed precision training to accelerate training. To install, please follow instruction here

    • You can remove apex dependency if
      • you have more powerful GPUs, or
      • you don't need to run the training script.
    • Note: We tried to use the native mixed precision training from official Pytorch implementation. However, it currently does not support gradient checkpointing for LayerNorm. We will post update if this is resolved. ## Pre-trained Models You can download the pretrained model from the following links.
  • Google Drive:

| Models | Link | |:-------------------------- |:---------:| | STTR (Scene Flow pretrained) | Download link | | STTR (KITTI finetuned) | Download link | | STTR-light (Scene Flow pretrained) | Download link |

  • Baidu download link:
    • Link:
    • Password:

Folder Structure

Code Structure

    |_ dataset (dataloder)
    |_ module (network modules, including loss)
    |_ utilities (training, evaluation, inference, logger etc.)

Data Structure

Please see sample_data folder for details. We keep the original data folder structure from the official site. If you need to modify the existing structure, make sure to modify the dataloader.

  • Note: We only provide one sample of each dataset to run the code. We do not own any copyright or credits of the data.

Scene Flow

    |_ RGB_finalpass
        |_ TRAIN
            |_ A
    |_ disparity
        |_ TRAIN
            |_ A
    |_ occlusion
        |_ TRAIN
            |_ left

MPI Sintel

    |_ training
        |_ disparities
        |_ final_left 
        |_ final_right 
        |_ occlusions (occlusions of left border of objects)
        |_ outofframe (occlusion of left border of images)

KITTI 2015

    |_ training
        |_ disp_occ_0 (disparity including occluded region)
        |_ image_2 (left image)
        |_ image_3 (right image)

MIDDLEBURY2014 ``` MIDDLEBURY2014 |_ trainingQ |_ Motorcycle (scene name) |_ disp0GT.pfm (left disparity) |_ disp1GT.pfm (right disparity) |_ im0.png (left image) |_ im1.png (right image) |_ mask0nocc.png (left occlusion) |_ mask1nocc.png (right occlusion) ```


    |_ training
        |_ disp_left
        |_ img_left 
        |_ img_right
        |_ occ_left 


Colab/Notebook Example

If you don't have a GPU, you can use Google Colab: - An example of how to run inference is given in the Colab example

If you have a GPU and want to run locally: - Download pretrained model using links in Pre-trained Models. - Note: The pretrained model is assumed to be in the

folder. - An example of how to run inference is given in file inference_example.ipynb.

Terminal Example

  • Download pretrained model using links in Pre-trained Models.
  • Run pretraining by
    sh scripts/
    • Note: please set the
      argument in the
      file to where Scene Flow data is stored, i.e. replace
  • Run fine-tune on KITTI by
    sh scripts/
    • Note: please set the
      argument in the
      file to where KITTI data is stored, i.e. replace
    • Note: the pretrained model is assumed to be in the
  • Run evaluation on the provided KITTI example by
    sh scripts/
    • Note: the pretrained model is assumed to be in the

Expected Result

The result of STTR may vary by a small fraction depending on the trial, but it should be approximately the same as the tables below.

Expected result of STTR (

) and STTR-light (


| | 3px Error | EPE | Occ IOU | |:----------: |:---------------: |:---------------: |:---------------:| |STTR | 1.26 | 0.45 | 0.92 | |STTR-light | 1.54 | 0.50 | 0.97 |

Generalization without fine-tuning.

| | MPI Sintel | | | KITTI 2015 | | | Middleburry-Q | | | SCARED | | | |:----------: |:---------------: |:---------------: |:---------------:|:---------------:|:---------------:|:---------------:|:---------------: |:--------------: |:---------------: |:------: |:-----------------: |:-----------------: | | | 3px Error | EPE | Occ IOU | 3px Error | EPE |Occ IOU | 3px Error | EPE | Occ IOU| 3px Error | EPE |Occ IOU | 3px Error | EPE | Occ IOU | |STTR | 5.75 | 3.01 | 0.86 | 6.74 | 1.50 | 0.98 | 6.19 | 2.33 | 0.95 | 3.69 | 1.57 | 0.96 |STTR-light | 5.82 | 2.95| 0.69 | 7.20 | 1.56 | 0.95 | 5.36 | 2.05 | 0.76 | 3.30 | 1.19 | 0.89

Expected 3px error result of



3px Error EPE
KITTI 2015 training 0.79 0.41
KITTI 2015 testing 2.01 N/A

Common Q&A

  1. I don't see occlusion from Scene Flow dataset. What should I do?\ Scene Flow dataset can be downloaded at However, you may notice that the Full datasets has disparity and images, but not occlusion. What you need to do is to download the occlusion from the DispNet/FlowNet2.0 dataset subsets and use the provided training list on the right named train to only use the subset of Full datasets with the occlusion data. Please see

    for more details of subsampling the Full datasets.
  2. How much memory does it require to train/inference?\ We provide a flexible design to accommodate different hardware settings.

    • For both training and inference, you change the
      parameter to reduce memory consumption at the cost of potential performance degradation.
    • For training, you can always change the crop size in
    • For both training and inference, you can use the light-weight model STTR-light.
  3. What are occluded regions?\ "Occlusion" means pixels in the left image do not have a corresponding match in right images. Because objects in right image are shifted to the left a little bit compared to the right image, thus pixels in the following two regions generally do not have a match:

    • At the left border of the left image
    • At the left border of foreground objects
  4. Why there are black patches in predicted disparity with values 0?\ The disparities of occluded region are set to 0.

  5. Why do you read disparity map including occluded area for KITTI during training?\ We use random crop as a form of augmentation, thus we need to recompute occluded regions again. The code for computing occluded area can be found in dataset/

  6. How to reproduce feature map visualization in Figure 4 of the paper?\ The feature map is taken after the first LayerNorm in Transformer. We use PCA trained on the first and third layer to reduce the dimensionality to 3.


This project is under the Apache 2.0 license. Please see LICENSE for more information.


We try out best to make our work easy to transfer. If you see any issues, feel free to fork the repo and start a pull request.


Special thanks to authors of SuperGlue, PSMNet and DETR for open-sourcing the code. We also thank GwcNet, GANet, Bi3D, AANet for open-sourcing the code. We thank Xiran for MICCAI pre-processing.

We use cookies. If you continue to browse the site, you agree to the use of cookies. For more information on our use of cookies please see our Privacy Policy.