No description, website, or topics provided.
Switch branches/tags
Nothing to show
Clone or download
Latest commit 7de562f Oct 20, 2018
Failed to load latest commit information.
odes "main code" Oct 20, 2018
scripts "main code" Oct 20, 2018
wavedata "main code" Oct 20, 2018
.gitignore ignore Oct 20, 2018
LICENSE "main code" Oct 20, 2018 Update Oct 20, 2018 "main code" Oct 20, 2018 "main code" Oct 20, 2018 "main code" Oct 20, 2018
requirements.txt "a" Oct 20, 2018 "main code" Oct 20, 2018


The paper is still underview, and description will be released here after it is accepted.


Implemented and tested on Ubuntu 16.04 with Python 3.5 and Tensorflow 1.8.0.

  1. Clone this repo

    >>git clone --recurse-submodules
  2. Install tensorflow-1.8.0

  3. Install Python dependencies

    >>cd Object_detection_estimated_sclales
    >>pip3 install -r requirements.txt
  4. Compile integral image library in wavedata

    >>sh scripts/install/build_integral_image_lib.bash
  5. Protobufs are used to configure model and training parameters. Before the framework can be used, the protos must be compiled:

    >>sh odes/protos/


  1. Download Kitti Object Detection Dataset, and place it in your home folder at ~/Kitti/object

  2. Split data (split_ratio = 0.75 in our setting)

    >>python odes/utils/

    The train.txt and val.txt will be generated, and copy them into ~/Kitti/object. If you want to try other split ratio, adjust split_ratio in

  3. Download planes into ~/Kitti/object

  4. Download the pretrained model VGG16 in tensorflow models, and unzip the pretained model into odes/data/pretrained_models


  1. Generate preprocessed data including estimated scale size from depth.

    >>chmod +x

    The configuration file can be found under directory ./odes/configs/mb_preprocessing/. Changing parameter settings like area size, density threshold etc. in the config file according to your application.

  2. Start to train.

    >>chmod +x

    The training configuration can be found under directory ./odes/configs/ where you can find the max number of iterations, learning rate, optimizer and pretrained model dir etc.


  1. Run evaluation code

    >>chmod +x

    If you have the multiple GPU, you can run trianing and evluating simultaneously on different GUPs by setting CUDA_VISIBLE_DEVICES in the file.

  2. Evaluting results.

    AP Easy Moderate Hard
    Car 93.53 90.89 90.62
    Pedestrain 80.39 74.85 66.76
    Cyclist 81.33 76.73 71.47

    Evaluate your model around 380000 Iterations, you should be able to achieve above accuracy


Thanks to the team of Jason Ku , we have benifited a lot from their previous work AVOD (Aggregate View Object Detection) and his code.