Purpose

To estimate distance to objects (cars, pedestrians, trucks) in the scene on the basis of detection information

Overview

Train a deep learning model that takes in bounding box coordinates of the detected object and estimates distance to the object.

Input: bounding box coordinates (xmin, ymin, xmax, ymax)
Output: distance (z)

Usage

To train and test the models, execute the following from distance-estimator directory, unless mentioned otherwise

Training

1. (Optional) Use hyperopti.py for hyperparameter optimization. Choose the hyperparameters you would like to try out. (Default model inside hyperopti trains on two gpus, change it if you want.) More info on hyperoptimization here
2. You can use result of 1. and edit train.py accordingly. Otherwise, use train.py to define your own model, choose hyperparameters, and start training!

Inference

1. Use inference.py to generate predictions for the test set.

python inference.py --modelname=generated_files/[email protected] --weights=generated_files/[email protected]

2. Use prediction-visualizer.py to visualize the predictions.

cd KITTI-distance-estimation/
python prediction-visualizer.py

Results

Appendix

Prepare Data

1. Download KITTI dataset

# get images
wget https://s3.eu-central-1.amazonaws.com/avg-kitti/data_object_image_2.zip
unzip data_object_image_2.zip

# get annotations
wget https://s3.eu-central-1.amazonaws.com/avg-kitti/data_object_label_2.zip
unzip data_object_label_2.zip

Organize the data as follows:

KITTI-distance-estimation
|-- original_data
    |-- test_images
    |-- train_annots
    `-- train_images

2. Convert annotations from .txt to .csv
   We only have train_annots. Put all information in the .txts in a .csv

python generate-csv.py --input=original_data/train_annots --output=annotations.csv

The annotations contain the following information

Values    Name      Description
----------------------------------------------------------------------------
   1    type         Describes the type of object: 'Car', 'Van', 'Truck',
                     'Pedestrian', 'Person_sitting', 'Cyclist', 'Tram',
                     'Misc' or 'DontCare'
   1    truncated    Float from 0 (non-truncated) to 1 (truncated), where
                     truncated refers to the object leaving image boundaries
   1    occluded     Integer (0,1,2,3) indicating occlusion state:
                     0 = fully visible, 1 = partly occluded
                     2 = largely occluded, 3 = unknown
   1    alpha        Observation angle of object, ranging [-pi..pi]
   4    bbox         2D bounding box of object in the image (0-based index):
                     contains left, top, right, bottom pixel coordinates
   3    dimensions   3D object dimensions: height, width, length (in meters)
   3    location     3D object location x,y,z in camera coordinates (in meters)
   1    rotation_y   Rotation ry around Y-axis in camera coordinates [-pi..pi]
   1    score        Only for results: Float, indicating confidence in
                     detection, needed for p/r curves, higher is better.

3. Generate dataset for distance estimation
   Using only annotations.csv (file generated using train_annots), split the dataset into train.csv and test.csv set.

python generate-depth-annotations.py

This dataset contains the following information: filename, xmin, ymin, xmax, ymax, angle, xloc, yloc, zloc

Organize your data as follows

KITTI-distance-estimation
|-- original_data
|    |-- test_images
|    |-- train_annots
|    `-- train_images
`-- distance-estimator/
    |-- data
        |-- test.csv
        `-- train.csv

4. Visualize the dataset
   Use visualizer.py to visualize and debug your dataset. Edit visualizer.py as you want to visualize whatever data you want.

Training

1. Use hyperopti.py for hyperparameter optimization. Choose the hyperparameters you would like to try out. More info on hyperoptimization here
2. Use result of 1. and edit train.py accordingly. Use train.py to actually train your model
3. Use inference.py to generate predictions for the test set.
4. Use prediction-visualizer.py to visualize the predictions.

TODO

1. Save models in hyperopti.py so train.py wont be necessary (waiting on hyperas issue)
2. Handle num_gpus (cannot access global variables inside create_model)

Acknowledgements

KITTI Vision Benchmark Suite